Home

Glide-related material 1. Richard A. Friesner, Jay L. Banks, Robert B

1. Figure 3 2 The Ligands tab of the Ligand Docking panel 3 2 Specifying Ligands To Dock There are several methods for specifying ligand structures to be docked with receptor grids In this tutorial you will specify a file containing a set of 50 ligands 1 In the Ligands tab ensure that File is selected 2 Click Browse A file selector is displayed Ensure that Files of type is set to Maestro 3 Navigate to the tutorial structures directory choose 50ligs mae gz and click Open 4 Ensure that the selected Range is from 1 to End the default 5 Ensure that van der Waals radii scaling for ligand atoms is set to the default values Scal ing factor to 0 80 and Partial charge cutoff to 0 15 Glide 5 5 Quick Start Guide 17 Chapter 3 Ligand Docking 18 In this docking job the constraints that were specified in the grid generation will not be used In a later exercise you will use the constraints to dock the same ligands 3 3 Specifying Output Quantity and File Type The Output tab allows you to specify the type of file to create for the output ligand poses and to determine how many poses to write per ligand and per docking job 1 In the Output tab ensure that Write pose viewer file includes receptor filename will be lt jobname gt _pv mae is selected You are specifying that the structural output from the docking job be written to a pose viewer file a file of ligand poses that begins with the struct
2. Hydrogen bonds visualization 1 Macromolecule preparation WORKFLOWS Protein Preparation Wizard gt Preprocess Assign bond orders Add hydrogens Delete waters etc gt Optimize Minimize ALWAYS check Job Monitor Status 2 Ligand preparation Change bond order gt Edit gt Atom Properties gt Atom type Add hydrogens H Edit gt Clean up geometry LigPrep generate possible states at target pH 7 lonizer gt Start 3 GLIDE gt Receptor Grid generation Site gt Centroid of selected residues Constraints gt H bond Rotatable groups gt Start 4 GLIDE gt Ligand Docking gt Receptor Grid name SP Dock flexibly Ligands workspace Constraints Output number poses per ligand write per residue interactions scores compute RMSD write report Start incorporate new entries as a new group 5 Analysis Open poseviewer file name_pv maegz View table and setup Pose display Entry gt View Poses gt Setup Compare docking results with native solutions
3. Figure 2 3 The marked ligand with enclosing box 2 4 Setting Up Glide Constraints The Constraints tab of the Receptor Grid Generation panel is used to define Glide constraints In this exercise you will define two constraints a positional constraint and an H bond constraint To make it easier to see the parts of the receptor close to the ligand and to see the ligand atoms you will first change the display For more information on using Glide constraints see Section 4 4 of the Glide User Manual 2 4 4 Setting the Display for Constraint Definition 1 From the Display only selected atoms toolbar button menu choose Molecules and click on a ligand atom ilt The ligand is displayed and the receptor remains displayed as ribbons 2 From the Show hide or color ribbons button menu choose Delete Ribbons e 3 From Display residues within N of currently displayed atoms choose 3 an ms iei a L a Glide 5 5 Quick Start Guide 9 Chapter 2 Receptor Grid Generation 10 The residues that are closest to the ligand are displayed From the Undisplay toolbar button menu choose Nonpolar hydrogens lt j This action leaves the polar hydrogens displayed making it easier to see the polar hydro gens that are likely to form hydrogen bonds 2 4 2 Defining a Positional Constraint In the Constraints tab of the Receptor Grid Generation panel click the Positional tab Click New The New Po
4. Delete All X Show markers Label positions Start Write Reset Close Figure 2 5 The Constraints tab of the Receptor Grid Generation panel showing the Positional subtab 2 4 3 Defining an H bond Constraint Next you will define an H bond constraint for the carboxylate that is H bonded to the amidine of the ligand To aid the picking of the constraint H bonds to the ligand will be displayed 1 From the Display H bonds button menu choose Inter H bonds and click on a ligand atom LI H The hydrogen bonds between the ligand and the receptor appear as yellow dashed lines Glide 5 5 Quick Start Guide 11 Chapter 2 Receptor Grid Generation 12 Receptor Grid Generation Receptor Site Constraints Rotatable Groups e mn x 2 constraints have been defined limit is 10 total Positional 1 H bond Metal 1 Hydrophobic 0 Pick receptor atoms that could participate in hydrogen bond or metal ligand interactions during docking Ligand interactions with these atoms may be chosen as constraints during docking Receptor atoms 1484 A ASP 189 0D2 1484 0D2 1483 0D1 Gl a Delete All Pick atoms Show markers Label atoms Start Write Reset Close Help Figure 2 6 The Constraints tab of the Recepto
5. The Receptor Grid Generation Start dialog box is displayed 2 In the Output section ensure that the Directory for grid files is the default your cur rent working directory and ensure that Compress is selected When this option is selected the grid files are placed in an archive that is compressed so it is easy to copy Glide can make use of this compressed archive directly and extracts the files from it as needed 3 Check the main window title bar to confirm that the current working directory is your path tutorial grids 7 Receptor Grid Generation Start Output Directory for grid files Browse L Compress Job Standard names glide grid 1 IN Name glide grid 1 Host localhost 1 v Username Entry Title Figure 2 7 The Receptor Grid Generation Start dialog box Glide 5 5 Quick Start Guide 13 Chapter 2 Receptor Grid Generation 14 4 In the Job section change the Name to factorXa grid 5 Choose a host and if necessary specify a user name 6 Start the job by clicking Start A warning dialog box might be displayed informing you that the structure has not been prepared with the Protein Preparation Wizard You can ignore this warning and click Continue After a moment the Monitor panel is displayed and the job starts While the job is in progress the Status column displays running When the jo
6. sis also produced trial values for the coefficients of the various penalty terms For the most part these coef ficients were used without modification in GlideScore 2 5 XP For Glide 2 5 SP however the need to make the program relatively fast limits the amount of sam pling that can be done during docking and hence limits the accuracy of the docked poses As a result optimiza tion of the solvation penalties led to smaller coefficients that do not too heavily penalize misdocked actives E ven these smaller values however provide substantial enhancement in enrichment factors for many database screens thrombin being the most prominent example in this case primarily by rejecting false positives that bury charged groups in a hydrophobic region of the thrombin active site 4 Docking Accuracy This section characterizes Glide s performance in reproducing the geometries of cocrystallized ligands taken from an extensive set of 282 publically available PDB complexes This set includes most of the members of the well known GOLD and FlexX test sets ap proximately 50 PDB complexes used in evaluations of Glide by prospective customers and approximately 50 more complexes whose experimental binding affinities have been used to develop one or more of the empirical scoring functions described in the literature e g Chem Score We used the latter complexes and others induded in the GOLD and FlexX test sets to calibrate the GlideScore
7. 2002 322 339 355 37 Perola E Charifson P S Conformational Analysis of Drug Like Molecules Bound to Proteins An Extensive Study of Ligand Reorganization upon Binding J Med Chem 2004 47 2499 2510 38 Halgren T A MMFF VI MMFF94s Option for Energy Minimiza tion Studies J Comput Chem 1999 20 720 729 39 Cho A E Guallar V Berne B J Friesner R A Importance of Accurate Charges in Molecular Docking Quantum Mechanical Molecular Mechanical QM MM Approach J Comput Chem 2005 26 915 931 40 Schr dinger User Manuals Glide v3 0 Schr dinger L L C New York NY 1994 41 QikProp v2 3 Schr dinger L L C New York NY 2005 42 Wang R Fang X Lu Y Wang S The PDBbind Database Collection of Binding Affinities for Protein Ligand Complexes with Known Three Dimensional Structures J Med Chem 2004 47 2977 2980 JM0512560 Glide 5 5 Quick Start Guide Q Schr dinger Press Glide Quick Start Guide Copyright 2009 Schr dinger LLC All rights reserved While care has been taken in the preparation of this publication Schr dinger assumes no responsibility for errors or omissions or for damages resulting from the use of the information contained herein Canvas CombiGlide ConfGen Epik Glide Impact Jaguar Liaison LigPrep Maestro Phase Prime PrimeX QikProp QikFit QikSim QSite SiteMap Strike and WaterMap are trademarks of Schr dinger
8. 3 ChemScore does not treat desolvation effects 4 ChemScore uses a simple rotatable bond term to treat conformation entropy effects arising from restricted motion of the ligand The new XP Glide scoring function starts from the standard terms discussed above though the functional form of the first three terms have been significantly revised and the parameteriza tion of all terms is specific to our scoring function In the remain der of this section the functional form and physical rationale for the novel scoring terms we have developed are described with examples from pharmaceutically relevant test cases pro vided to illustrate how the various terms arise from consideration of the underlying physical theory and experimental data Form of the XP Glide Scoring Function The XP Glide scoring function is presented in eq 2 The principal terms that favor binding are presented in eq 3 while those that hinder binding are presented in eq 4 A description of each of the following terms besides Em pair and Ephobic pair Which are standard ChemScore like hydrogen bond and lipophilic pair terms respectively follows XP GlideScore Esou Eyaw Ebina Epenatty 2 coul Evina E hyd enclosure b Eib nn motif Ey cc motif 25 Eg Erp pair E bashes puit 3 E vindi Ead T E igang stran 4 Improved Model of Hydrophobic Interactions Hydro phobic Enclosure Enya_enclosure The ChemScore atom atom pair function Ephobic_pai
9. 8 1 6 3 1 46 1 11 4aah 7 6 7 6 11 0 0 25 0 24 lphg 11 8 10 9 10 9 8 1 4 29 1 20 Acts 4 6 4 6 8 8 0 25 0 19 lpoc 10 7 8 4 8 4 10 1 1 44 1 52 Adfr 11 8 10 1 10 1 9 3 0 92 5 17 lppc 84 11 9 9 9 9 8 1 40 6 31 4fab 11 0 13 7 12 9 9 1 4 44 0 82 lpph 8 1 10 5 8 5 10 4 0 70 0 58 4fbp 9 7 9 7 12 1 2 03 0 55 lppi 16 7 14 7 8 5 1 01 2 80 4fxn 173 173 132 0 50 0 49 lppk 10 4 9 9 9 9 10 1 0 73 0 27 4hmg 3 5 6 6 6 6 6 7 0 54 0 67 Ippl 11 7 10 3 10 3 11 6 0 70 2 55 4phv 125 14 5 14 5 11 8 0 55 4 22 lppm 7 9 12 5 12 5 117 0 99 0 62 4tim 2 9 2 3 5 3 10 2 1 32 1 32 lpro 15 4 15 6 15 6 12 7 1 50 1 51 4tln 5 1 l 7A1 6 6 1 43 2 67 lpso 14 1 10 9 10 9 9 6 5 15 6 12 4tmn 13 9 11 9 11 9 12 1 1 29 0 73 1sbg 10 6 11 1 H 10 9 0 88 0 40 4tpi 4 0 6 6 6 6 8 5 0 88 0 56 1slt 5 8 5 8 6 2 1 03 0 57 4ts1 6 7 8 6 8 6 8 8 0 89 0 85 Isnc 9 1 9 6 9 6 9 8 2 06 1 12 Sabp 9 1 1 6 1 6 8 5 0 11 0 20 lsre 5 3 8 9 8 5 10 0 0 30 0 36 Scpp 8 0 8 2 8 2 6 8 0 11 2 65 lsrj 13 7 12 9 10 0 0 47 0 49 5cts 3 2 32 8 1 0 27 0 27 Istp 18 3 18 2 18 2 10 7 0 60 0 58 5p2p 9 0 9 0 5 4 4 95 6 18 1tdb 1 9 79 7 6 7 34 7 50 5tim 3 2 3 2 ENa 1 32 0 69 Ithy 7 0 7 0 1 98 4 21 5tln 8 7 12 4 12 4 9 9 2 37 1 01 I
10. A file selector opens 4 Navigate to the tutorial grids directory choose factorXa_grid zip and click Open Glide 5 5 Quick Start Guide 15 Chapter 3 Ligand Docking 16 7 Ligand Docking EEE Settings Ligands Core Constraints Similarity Output Receptor grid Specify the receptor grid you want to use for docking Receptor grid base name 21 dyall glide 2008u1 g rids factorXa_grid Docking Precision O HTVS high throughput virtual screening SP standard precision XP extra precision Options Dock flexibly X Sample ring conformations Amide bonds Penalize nonplanar conformation Dock rigidly Refine do not dock Score in place do not dock _ Write XP descriptor information _ Add Epik state penalties to docking score Advanced Settings Start Write Reset Close Help Figure 3 1 The Settings tab of the Ligand Docking panel The Receptor grid base name is fullpath tutorial grids factorXa grid 5 In the Docking section ensure that the Precision option is SP standard precision This is usually the best choice for docking large numbers of ligands For more rapid screening you can use the HTVS high throughput virtual screening option You will do this in a later exercise 6 Under Options ensure that Dock flexibly and Sample Ring Conformations are selected and Penalize nonplan
11. E mail rich chem columbia edu Schr dinger L L C NY Schr dinger L L C OR 10 1021 jm0512560 CCC 33 50 van der Waals potential of the receptor conformation used in docking must be built into the potential energy function employed to predict the ligand binding mode In XP and SP Glide this is accomplished by scaling the van der Waals radii of nonpolar protein and or ligand atoms scaling the vdW radii effectively introduces a modest induced fit effect However it is clear that there are many cases in which a reasonable degree of scaling will not enable the ligand to be docked correctly For example a side chain in a rotamer state that is very different from that of the native protein ligand complex may block the ligand atoms from occupying their preferred location in the binding pocket There will always be borderline situations but in practice we have found it possible to classify the great majority of cases in cross docking experiments as either fitting or not fitting The former are expected to be properly ranked by XP Glide within the limitation of noise in the scoring function while the latter require an induced fit protocol to correctly assess their binding affinity In the present paper we focus on complexes where the ligand fits appropriately into the receptor as judged by two factors 1 the ability to make key hydrogen bonding and hydrophobic contacts and 2 the ability to achieve a rea
12. Enrichment factors in database screening J Med Chem 2004 47 1750 1759 17 J orgensen W L Maxwell D S Tirado Rives J Development and testing of the OPLS all atom force field on conformational energetics and properties of organic liquids J Am Chem Soc 1996 118 11225 11236 2 gt 3 4 6 S 7 8 J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1749 18 Eldridge M D Murray C W Auton T R Paolini G V Mee R P Empirical scoring functions 1 The development of a fast empirical scoring function to estimate the binding affinity of ligands in receptor complexes J Comput Aided Mol Des 1997 11 425 445 19 Murphy R B Friesner R A Halgren T A Unpublished research 20 Babine R E Bender S L Molecular recognition of protein ligand complexes Applications to drug design Chem Rev 1997 97 1359 1472 21 Whittaker M Floyd C D Brown P Gearing A J H Design and therapeutic application of matrix metalloproteinase inhibi tors Chem Rev 1999 99 2735 2776 22 Bell I M Gallicchio S N Abrams M Beese L S Beshore D C Bhimnathwala H Bogusky M J Buser C A Culber son J C Davide J Ellis Hutchings M Fernandes C Gibbs J B Graham S L Hamilton K A Hartman G D Heim brook D C Homnick C F Huber H E Huff J R Kassahun K et al 3 Aminopyrrolidino
13. Imdr 5 4 82 8 2 9 6 1 95 0 54 2lgs 34 34 95 0 88 0 53 1mfe 7 2 8 0 8 0 7 4 6 09 1 78 2mcp F 5 6 5 6 5 6 1 54 1 25 1mld 6 6 6 6 10 6 0 25 0 22 2phh 6 4 7 9 79 8 7 0 47 0 41 1mmq 10 3 13 8 13 8 11 0 0 67 0 33 2pk4 5 9 8 8 6 8 5 8 0 65 0 85 Imnc 12 3 11 9 11 9 1 1 0 33 0 73 2plv 14 3 11 8 Sa 1 78 1 90 lmrg 1 8 1 8 1 8 0 15 0 12 2r04 8 0 10 6 8 1 8 9 1 44 0 75 Imrk 6 2 11 3 1 3 94 1 21 1 17 2107 10 9 84 8 7 0 92 0 67 Imup 9 3 6 6 2 4 50 4 05 2sim 4 7 1 l 7 1 10 5 0 82 0 94 Inco 10 6 10 4 10 4 121 0 60 0 33 2tmn 8 0 10 1 10 1 10 4 0 65 0 50 Inis 4 1 4 0 4 0 75 0 26 0 45 2tpi 5 9 8 6 6 6 9 5 0 26 1 15 Innb 72 5 5 8 7 1 39 0 25 2upj 14 2 ALI 11 1 10 7 3 24 2 69 Insc 4 1 5 7 5 7 10 5 0 66 0 24 2xis 79 4 9 4 9 S 2 05 2 40 Insd EA 6 3 6 3 9 0 0 74 0 22 2yhx 4 3 4 3 5 6 1 91 2 19 lodw 11 5 11 5 5 7 3 91 2 59 3cla 6 7 6 4 6 4 5 4 5 09 6 06 lokl 8 2 72 72 5 8 0 38 3 14 3dfr 14 0 15 3 33 1 3 0 51 0 70 Ipbd 11 4 11 4 9 9 0 32 0 26 3hvt 11 9 11 4 9 0 0 72 0 79 lpgp 7 8 6 3 6 3 8 8 2 23 1 83 3mth 5 4 54 5 9 1 23 5 62 lpha 122 122 8 8 1 04 0 52 3ptb 6 1 6 5 6 5 8 8 0 23 0 16 Iphd 7 4 7 4 6 5 1 13 0 30 3tpi 5 9 9 8 7 8 9 0 0 47 0 51 Iphf 6 0 8 2
14. S L Molecular recognition of protein ligand complexes Applications to drug design Chem Rev 1997 97 1359 1472 18 Bell I M Gallicchio S N Abrams M Beese L S Beshore D C Bhimnathwala H Bogusky M J Buser C A Culber son J C Davide J Ellis Hutchings M Fernandes C Gibbs J B Graham S L Hamilton K A Hartman G D Heim brook D C Homnick C F Huber H E Huff J R Kassahun K et al 3 Aminopyrrolidinone F arnesyltransferase inhibitors design of macrocydic compounds with improved pharmacoki netics and excellent cell potency J Med Chem 2002 45 2388 2409 19 Schulz Gasch T Stahl M Binding site characteristics in structure based virtual screening E valuation of current docking tools J Mol Model 2003 9 47 57 20 Ewing T J A Kuntz I D Critical evaluation of search algorithms for automated molecular docking and database screening J Comput Chem 1997 18 1175 1189 21 Lorber D M Shoichet B K Flexible ligand docking using conformational ensembles Protein Sci 1998 7 938 950 22 Eldridge M D Murray C W Auton T R Paolini G V Mee R P Empirical scoring functions I The development of a fast empirical scoring function to estimate the binding affinity of ligands in receptor complexes J Comput Aided Mol Des 1997 11 425 445 23 Shoichet B K Kuntz I D Matching chemistry and shape in mo
15. but in this case more competitive set with an average molecular weight of 360 the dl 360 dataset The property distributions of these databases were characterized in the preceding paper We believe these compounds to be representative of the chemical sample collections of pharmaceutical and biotechnology com panies As such they should provide a fair and strin gent test of the efficacy of the docking method E ach screen used 1000 database ligands and between 7 and 33 known actives All compounds considered have 20 or fewer rotatable bonds and 100 or fewer atoms Likethe database ligands the known binders were also MMFF94s optimized but in these cases we used un biased input geometries obtained via a MacroM odel conformational search as previously described for ligands taken from cocrystallized compl exes Glide s use of reduced atomic van der Waals radii to mimic minor readjustments of the protein these should be distinguished from the more substantial induced fit rearrangements modeled by the use of multiple receptor conformations is an important issuein the setup of the docking runs Glide currently supports uniform van der Waals scaling of the radii of nonpolar protein and or ligand atoms To characterize the performance that can be expected when Glide is run out of the box the principal results presented in this paper use default 1 0 protein scaling which means that the OPLS AA van der Waal vdW radii are not cha
16. of those employed in ScreenScore or in its FlexX or PLP predecessors 13 Thelargest component of the objective function in the optimization process was the ranking of active com pounds in a large and diverse suite of database screen ing tests as measured by enrichment factors computed as described in the following paper 6 The second component was the fit of the predicted binding affinities for a set of 125 PDB cocrystallized complexes to experi mentally measured values the rmsd achieved in this case is 2 2 kcal mol a reasonable result but one that can dearly be improved In the parametrization of GlideScore 1 8 and 2 0 we used decoy ligands assembled from cocrystallized PDB complexes and from a small portion of the Comprehen sive Medicinal Chemsitry database in the database screens However it became apparent through more extensi ve testing that these ligands which average only 290 in molecular weight are too small to make enough favorable interactions with the protein site to compete fairly with the known actives for our screens which average about 410 in molecular weight when the very large HIV protease ligands are excluded This disparity in molecular weight distorted both the apparent enrich ment factors in the database screens and the param eters we obtained from the optimization process A second set of database ligands used in the present work was provided recently by a pharmaceutical col league This set consists o
17. using the information below E mail help schrodinger com USPS Schr dinger 101 SW Main Street Suite 1300 Portland OR 97204 Phone 503 299 1150 Fax 503 299 4532 WWW http www schrodinger com FTP ftp ftp schrodinger com Generally e mail correspondence is best because you can send machine output if necessary When sending e mail messages please include the following information All relevant user input and machine output Glide purchaser company research institution or individual Primary Glide user Computer platform type Operating system with version number Glide version number Maestro version number mmshare version number On UNIX you can obtain the machine and system information listed above by entering the following command at a shell prompt SSCHRODINGER utilities postmortem This command generates a file named username host schrodinger tar gz which you should send to help schrodinger com If you have a job that failed enter the following command SSCHRODINGER utilities postmortem jobid Glide 5 5 Quick Start Guide Getting Help where jobid is the job ID of the failed job which you can find in the Monitor panel This command archives job information as well as the machine and system information and includes input and output files but not structure files If you have sensitive data in the job launch directory you should move those files to another location f
18. 6 14 3 8 6 estrogen receptor 3ert 88 4 79 8 66 9 35 0 35 0 35 0 14 0 8 0 estrogen receptor lerr 88 4 37 5 46 7 35 0 30 0 30 0 14 0 9 0 CDK 2 kinase 1dm2 3 6 3 9 6 8 5 0 10 0 15 0 10 0 6 0 CDK 2 kinase laql 2 1 3 8 5 4 5 0 15 0 10 0 8 0 6 0 p38 MAP kinase 1a9u 2 0 1 8 2 5 0 0 2 9 5 9 4 7 2 4 p38 MAP kinase 1bl7 1 8 2 9 35 2 9 5 9 8 8 3 5 4 1 p38 MAP kinase 1kv2 4 5 2 9 4 8 8 8 11 8 8 8 8 2 4 1 HIV protease Ihpx 10 8 7 8 47 6 30 0 13 3 36 7 17 3 8 7 thrombin 1dwc 5 8 2 8 12 1 6 2 3 1 15 6 11 2 8 1 thrombin lett 5 2 5 6 38 0 12 5 3 1 34 4 16 2 9 4 thermolysin ltmn 1 6 15 2 24 5 5 0 15 0 25 0 18 0 9 0 Cox 2 1cx2 3 6 3 4 5 05 7 6 3 0 13 6 7 9 5 2 Cox 2 site 1 ligands 1cx2 5 7 5 7 13 9 10 9 4 3 19 6 11 3 7 4 HIV rev transcriptase Irt1 4 6 3 2 8 2 3 0 0 0 12 1 10 3 6 4 HIV RT lvrt 1 8 2 0 7 1 0 0 0 0 7 6 9 1 5 8 av enrichment factor 5 3 6 5 14 8 6 0 7 1 18 6 11 4 7 0 a Enrichment factors can be at most 50 for 2 sampling 20 for 5 sampling and 10 for 10 sampling P EF 60 value 100 80 60 40 20 Figure 1 Percent of actives recovered with Glide 2 5 for assaying 296 596 and 1096 of the ranked database for the screens considered in this paper The PDB codes are defined in Table 1 is appropriate because what is most needed is a method that can be counted on to always perform reasonably well rather than one that does very well for some systems but is useless for others Table 1 shows that Glide 2 5 is much b
19. 8 7 8 7 9 1 0 44 0 29 lhef 12 1 11 2 112 10 3 6 24 6 43 lctr 5 8 7 5 715 5 7 2 27 2 59 hfc 7 5 8 9 8 9 94 2 25 2 26 lett 6 2 6 9 6 9 7 0 0 60 5 04 lhgg 3 4 6 4 6 4 12 1 37 1 47 1d3d 12 4 12 7 12 7 10 4 1 52 2 74 Ihgh 3 9 5 7 S T 6 2 4 70 0 49 1d3p 8 9 9 6 9 6 10 3 1 72 2 05 Ihgi 3 7 3 9 3 9 6 9 0 37 0 23 1d7x 1 2 a 9 1 0 58 0 44 Ihgj 2 3 4 8 4 8 4 0 0 64 0 34 1d8f 6 8 6 8 6 8 4 25 4 31 Thih 11 0 11 8 11 8 11 0 1 26 1 29 1dbb 12 3 13 2 13 0 10 3 0 37 0 41 Ihps 12 6 122 122 12 8 11 93 2 09 1dbj 10 4 15 9 13 4 9 5 0 32 0 21 lhpv 12 6 9 3 9 3 10 0 1 05 0 93 ldbk 11 0 15 2 12 9 9 0 0 57 0 40 Ihpx 12 7 11 5 115 13 1 3 34 3 31 ldbm 12 9 15 3 13 1 9 2 1 95 1 97 Ihri 5 9 8 9 8 9 L3 10 09 2 26 1dd6 13 5 13 3 9 1 1 36 8 27 hsg 12 8 12 5 125 13 3 0 41 0 35 1dds 11 3 10 8 10 8 7 4 1 75 2 38 Ihsl 9 8 8 4 6 4 9 4 1 31 1 31 ldhf 10 1 8 7 8 7 8 5 6 31 5 62 Ihte 8 6 9 9 9 9 9 6 7 32 7 36 1did 4 8 3 8 3 8 6 6 3 27 4 15 Ihtf 9 3 10 9 10 9 9 4 2 19 2 74 Idie 2 9 5 8 5 8 6 9 0 34 0 77 Ihti 7 0 4 5 4 5 5 7 4 40 1 60 Idih 7 8 8 2 8 2 14 3 2 62 1 78 Ihvr 13 0 13 7 13 7 7 8 1 60 1 75 1dm2 13 9 13 9 10 3 0 66 0 69 Ihyt 8 0 8 0 10 7 2 65 0 43 1dog 5 5 6 5 6 5 9 5 3 41 3 417 lien 10 3 7 8 1 7 9 06 2 04 1drl 7 4 7
20. Chapter 3 Ligand Docking 22 Ligand Docking EEE Settings Ligands Core Constraints Similarity Output Select constraints to use in docking Constraints can be grouped each group of constraints must be satisfied Optional constraints can be defined within a group Total number of constraints requested 1 Maximum 4 L Display receptor X Show markers Group 1 1 required Group 2 0 required Group 3 0 required Group 4 Available constraints 1 in use 1S4 arom Positional Custom x Sl site H bond Donor s ag Edit Feature Must match C All At least 1 L Test constraint satisfaction only after docking Start Write Reset Close Help Figure 3 5 The Constraints tab of the Ligand Docking panel 3 7 Docking Ligands Using Constraints In this exercise you will apply the constraints you defined in the grid generation to the docking of the same set of ligands as for the standard SP job By default no constraints are applied even if they are defined Here you will require any one of the three constraints to be applied 1 In the Settings tab select SP standard precision The other settings will be left as they were for the HTVS docking job 2 In the Constraints tab click both check boxes in the Use column These check boxes mark the constraint for use in docking 3 Under Mus
21. LLC Schr dinger and MacroModel are registered trademarks of Schr dinger LLC MCPRO is a trademark of William L Jorgensen Desmond is a trademark of D E Shaw Research Desmond is used with the permission of D E Shaw Research All rights reserved This publication may contain the trademarks of other companies Schr dinger software includes software and libraries provided by third parties For details of the copyrights and terms and conditions associated with such included third party software see the Legal Notices for Third Party Software in your product installation at SSCHRODINGER docs html third party legal html Linux OS or SCHRODINGER docs htmi third_party_legal htmi Windows OS This publication may refer to other third party software not included in or with Schr dinger software such other third party software and provide links to third party Web sites linked sites References to such other third party software or linked sites do not constitute an endorsement by Schr dinger LLC Use of such other third party software and linked sites may be subject to third party license agreements and fees Schr dinger LLC and its affiliates have no responsibility or liability directly or indirectly for such other third party software and linked sites or for damage resulting from the use thereof Any warranties that we make regarding Schr dinger products and services do not apply to such other third party software or linked site
22. a full text search the FAQ pages the New Features pages and several other topics If you do not find the information you need in the Maestro help system check the following sources Maestro User Manual for detailed information on using Maestro Maestro Command Reference Manual for information on Maestro commands Maestro Overview for an overview of the main features of Maestro Maestro Tutorial for a tutorial introduction to basic Maestro features Glide User Manual for detailed information on using Glide Protein Preparation Guide for information on protein preparation for Glide Impact Command Reference Manual for information on Impact commands Glide 5 5 Quick Start Guide 35 Getting Help 36 Glide Frequently Asked Questions pages at https www schrodinger com Glide FAQ html Known Issues pages available on the Support Center The manuals are also available in PDF format from the Schr dinger Support Center Local copies of the FAQs and Known Issues pages can be viewed by opening the file Suite 2009 Index html which is in the docs directory of the software installation and following the links to the relevant index pages Information on available scripts can be found on the Script Center Information on available software updates can be obtained by choosing Check for Updates from the Maestro menu If you have questions that are not answered from any of the above sources contact Schr dinger
23. a name for the folder If you want to create a folder inside this folder open the folder and repeat steps 2 and 3 4 Open the folder that contains the tutorial files This folder is in the Schr dinger software installation which by default is installed at C Schrodinger2009 a Open an explorer window b Navigate to the Schr dinger software installation c Open the impact vversion folder version is the 5 digit Impact version number then open the tutorial folder inside that folder 5 Drag the structures folder to the folder you created in Step 3 You can close the tutorial folder now 6 Create folders named glide and grids in the working folder You can use Step 2 and Step 3 for this task 1 3 Starting Maestro and Setting the Working Directory Once you have created the working directory you can start Maestro and set the Maestro working directory By default Maestro writes job files to its working directory You can change the default in the Preferences panel If you have changed the default you should change it back for this tutorial Glide 5 5 Quick Start Guide 3 Chapter 1 Getting Started UNIX If you have followed the directions in the previous section the SCHRODINGER environment variable should be already set and you should be in the working directory You can then skip the first two steps 1 Set the SCHRODINGER environment variable to the installation directory csh tesh setenv SCHRODINGER instal
24. agreement with experimental data One can speculate that the lack of charge complementarity in charged neutral hydrogen bonding pre cludes such structures from being major molecular recognition motifs though further investigations with larger data sets will be needed to resolve this issue Special Neutral Neutral Hydrogen Bond Motifs Ehpb nn motit In this section neutral neutral hydrogen bonding motifs are described that were identified based on both theoretical and empirical considerations as making exceptional contributions to binding affinity Such special hydrogen bonds represent key molecular recognition motifs that are found in many if not most pharmaceutical targets Targeting such motifs is a central strategy in increasing the potency and specificity of medicinal compounds Identifying such motifs through their incorporation in the scoring function should enable a dramatic improvement in both qualitative and quantitative predictions The critical idea in our recognition of special hydrogen bonds is to locate positions in the active site cavity at which a water molecule forming a hydrogen bond to the protein would have particular difficulty in making its complement of additional hydrogen bonds Forming such a hydrogen bond imposes nontrivial geometrical constraints on the water molecule This is the basis for the default hydrogen bond score but such constraints become more problematic when the environment of the water m
25. as compared to SP Glide are on average significant For PPARy both methods do reasonably well this XP Glide Methodology and Application Journal of Medicinal Chemistry 2006 Vol 49 No 21 6195 Table 11 Standard Enrichment Factors for Recovering 40 70 and 100 of Known Active Ligands in the Training Set Including Misdocked Cases enrichment factors v4 0 XP v2 7 XP v4 0 SP screen 40 70 100 40 70 100 40 70 100 acetylcholinesterase 37 19 1 1 1 1 2 2 1 neuramididase 64 34 6 2 1 1 112 23 4 factor Xa 78 58 3 10 1 1 22 9 1 human p38 map kinase 81 59 4 58 11 9 51 21 3 p38 map kinase 27 14 1 13 3 1 3 2 2 HIV RT 27 19 3 12 12 0 11 6 0 cyclooxygenase 2 35 29 T 56 29 0 78 58 9 human cyclin dep kinase 81 20 2 51 4 2 5 3 2 thrombin 64 64 2 54 7 0 11 10 3 HIV 1 protease 32 32 14 27 16 2 72 60 1 human estrogen receptor 101 101 2 101 88 2 101 17 2 Ick kinase 9 8 2 4 3 1 EGRE tyrosine kinase 5 6 1 1 I 1 2 2 1 thermolysin 42 52 23 112 37 2 168 34 7 thymidine kinase 251 251 201 251 188 201 17 24 17 Table 12 Test Set Used to Validate XP Virtual Screening PDB no No well docked code description actives actives Im4h BACE TI 34 Idan factor VIla 93 40 1fm6 PPAR closed form 93 32 1fm9 PPARy open form 93 25 1y6b Vegfr2 closed form 111 21 lywn Vegfr2 open form 111 26 lagl human cyclin dep kinase 253 143 lett thrombin 40 15 a All correctly docked ligands have experimental acti
26. collected and written to the pose viewer file 3 Ensure that Perform post docking minimization is selected with the default number of poses per ligand which is 5 Post docking minimization in the field of the receptor produces better poses and only adds a small amount to the time taken 3 4 Setting Up Distributed Processing If you have access to a host machine with multiple CPUs Glide can divide your multiple ligand docking job into subjobs that can be distributed over several processors The Start dialog box allows you to specify the number of subjobs and the number of processors to use In this section it is assumed that a host with five or more processors is available If you have access to such a host you can follow the instructions in this section without alteration If the host has fewer than five processors or you do not want to use five of its processors for this job change the settings below accordingly If you cannot or do not want to distribute processing over multiple CPUs skip to Section 3 5 1 Click Start The Start dialog box opens 2 Change the value in the Separate docking into N subjobs text box to 10 The docking job will be split into 10 subjobs each one docking 5 ligands For the sake of speed the number of ligands per subjob in this exercise is much smaller than would be typical in actual use The default for N is 1 meaning that the job is run as a single job 3 Choose a multiprocessor host from the H
27. eliminates high energy conformers by evalu ating the torsional energy of the various minima using a truncated version of the OPLS AA molecular mechan ics potential function and by imposing a cutoff of the allowed value of the total conformational energy com pared tothe lowest energy state The parameters of the heuristic screening function were optimized via exten sive testing on cocrystallized PDB complexes as de scribed below Each ligand is divided into a core region and some number of rotamer groups Figure 2 each of which is attached to the core by a rotatable bond but does not itself contain additional rotatable bonds That is the J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1745 lt rotomer group em O MN ii o rotomer group Figure 2 Definition of core and rotamer groups The four central torsions are part of the core Note that methyl groups are not considered rotatable Table 8 Maximum Number of Conformers Allowed vs Number of Core Degrees of Freedom no of core degrees of max no of actual no of core conformers freedom conformers allowed kept for individual ligands 3 120 4 13 17 18 4 120 4 6 18 19 24 34 5 120 9 106 106 6 150 24 28 150 7 214 214 214 8 278 278 278 9 342 342 342 10 406 406 406 12 534 534 534 14 662 662 662 Number of rotatable bonds in the core plus number of conformati onally labile five and six
28. energy scale Larger RMS values are indicated with a in cases where a nearly chemically symmetric solution was found by XP docking The pdbbind dataset v2002 lists a binding affinity of 13 6 kcal mol but this is for the entire antigen antibody complex whereas the structure provided is for a fragment of the complex that appears to lack essential antigen antibody interactions affinities This idea could form the basis for an approach enabling scores achieved in different protein conformations for a given receptor to be related based on experimental calibration of reorganization energy Performance of the methodology when there is substantial reorganization is addressed in a preliminary fashion by the enrichment studies in section 4 where a number of such ligand receptor pairs are considered In these cases knowledge of the reorganization energy of the receptor is not necessary to rank order the binding of compounds to a particular form of the receptor To address less extreme yet still nontrivial reorganization effects present in our data set we define average adjustable parameters to convert calculated empirical scores into predicted experimental binding affinities In particular we estimate the 6190 Journal of Medicinal Chemistry 2006 Vol 49 No 21 Table 6 RMS and Average Absolute Deviations Avg in Predicted Binding Affinities for the XP 4 0 and SP 4 0 Scoring Functions XP 4 0 SP 4 0 comparison numbe
29. following nine receptors for our initial studies five of which are represented by two or more alternative cocrystallized receptor sites thymidine kinase 1kim estrogen receptor 3ert lerr CDK 2 kinase 1dm2 1aq1 p38 MAP kinase 1a9u 1bl7 1kv2 HIV protease 1hpx thrombin 1dwc lett thermolysin 1tmn Cox 2 1cx2 HIV RT 1vrt 1rt1 VONOUPWNH 10 1021 jm030644s CCC 27 50 2004 American Chemical Society Published on Web 02 27 2004 Enrichment Factors in Database Screening The receptors for these screens cover a wide range of receptor types and therefore provide a proper test of a docking method All were prepared using the procedure described in the preceding paper or an earlier version of that procedure The known binders for the first two systems were specified by Rognan and co workers while ligands for the CDK 2 kinase receptor screens and for p38 MAP kinase were provided by pharmaceutical and biotech collaborators For thrombin 12 of the 16 known binders were taken from the studies by Engh et al and by von der Saal et al Others are ligands for the same target protein taken from our docking accuracy test set or were developed from multitple sources in the literature As database ligands we employed druglike decoys that averaged 400 in molecular weight the dl 400 dataset in most cases For thymidine kinase 1kim which has a very small active site however we used a similar
30. function Our coverage of the GOLD and FlexX sets is not quite complete because Glide does not deal with covalently attached ligands seven cases laec lase 1blh 1tpp 1lmp 3gch and 4est and cannot handle ligands having more than 35 rotatable bonds one case 2er6 In addition we exduded one complex 6rsa because it contains an atomic species vanadium for which the OPLS AA force field used by Glide has no parameters All results were obtained with release 2 5021 of the FirstDiscovery suite on an AMD Athelon MP 1800 processor running Linux All structures were prepared using the protein preparation procedure described in section 6 or an earlier version of that procedure For these calculations the vdW radii of nonpolar protein atoms were not scaled but the radii of nonpolar ligand atoms taken to be atoms having a partial charge of less than 0 15 e in magnitude were scaled down by a factor of 0 8 the same default scaling is also employed in the database enrichment studies presented in the accom panying paper 16 Friesner amp al Table 2 Average rms Deviations for Flexible Docking on 282 PDB Complexes no of no of av rms av CPU rotatable bonds cases top ranked pose time min 0 3 51 1 01 0 2 4 6 92 1 64 0 6 7 10 48 1 79 1 7 0 8 164 1 48 0 5 0 10 191 1 51 0 8 0 20 263 1 89 2 4 a Times are AMD Athelon MP 1800 CPU minutes To examine the dependence of the results on the starting geometries w
31. in the following paper 16 are very encouraging Furthermore database enrichment factors obtained using Glide 2 5 are significantly higher than those obtained using previous versions of Glide The paper is organized as follows Section 2 sum marizes the computational methodology used by Glide while section 3 describes Glide s approach to scoring relative ligand binding affinities The fourth section 10 1021 jm0306430 CCC 27 50 2004 American Chemical Society Published on Web 02 27 2004 1740 J ournal of Medicinal Chemistry 2004 Vol 47 No 7 Ligand conformations 1 Site point search N 2a Diameter test a N 2b Subsettost Nas Greedy score 2d Refinement _ 3 Grid minimization w Monte Carlo 4 Final scoring GlideScore Y Top hits Figure 1 Glide docking funnel showing the Glide docking hierarchy then presents rmsd values obtained for redocking co crystallized ligands and compares Glide to GOLD FlexX and Surflex methods we believe to be represen tative of the current state of the art in high throughput docking The fifth section summarizes the results and discusses future directions Finally section 6 provides details of the docking methodology and of the opti mization of the scoring function this section also de scribes the procedure we recommend for protein prep aration which in many cases can substantially affect the quality of the results obtained in docking calcula tions 2 O
32. leed 5 90 12 43 9 78 lela 160 n a 9 71 lelb 4 40 n a 7 17 lelc 8 22 n a 4 74 leld 0 67 n a 6 98 lele 2 52 n a 10 73 lepb 178 2 08 2 77 leta 2 92 1121 846 letr 148 423 7 24 lfen 0 66 n a 1 39 lfkg 125 181 7 59 Ifki 192 0 1 059 lfrp 0 27 n a 1 89 1ghb 189 145 133 1glp 0 34 n a 0 47 Iglq 029 135 643 Ihdc 058 10 49 11 74 Ihdy 174 0 94 n a Ihef 5 30 187 15 32 Ihfc 2 24 n a 2 51 lhgg 2 10 n a 10 05 Ihgh 0 28 n a 4 14 Ihgi 0 28 n a 0 97 Ihgj 0 18 n a 3 98 Ihri 159 14 01 10 23 1hsl 131 0 97 0 59 1hti 4 40 n a 1 54 Ihvr 150 n a 3 35 lhyt 028 110 162 licn 2 34 8 63 10 522 lida 11 88 12 12 11 95 ligj 130 942 717 limb 089 n a 4 71 livb 4 97 n a 1 29 livc 194 n a 2 21 livd 0 72 n a 5 42 live 261 216 534 livf 0 53 n a 6 97 lah 0 13 n a 0 28 1lcp 198 n a 1 65 idm 030 1 00 0 74 llic 4 87 10 78 5 07 1llmo 0 93 n a 4 49 Ina 0 95 n a 5 40 list 014 087 0 71 1mbi 1 668 n a 0 47 Ime 433 623 10 04 Imdr 052 0 36 0 88 1mid 032 n a 1 45 lmmq 0 92 n a 0 52 Imrg 0 30 n a 0 81 Imrk 120 101 3 55 Imup 4 37 396 3 8 1nco 6 99 n a 5 85 Inis 097 429 1 41 Insc 1 21 n a 2 12 lpbd 021 057 033 lpha 0 69 124 n a Iphd 122 085 0 65 1phf 114 n a 4 23 lphg 432 135 4 74 1poc 5 09 127 925 lppc 7 92 n a 3 05 lpph 4 31 n a 4 91 1ppi 6 24 n a 6 91 lppk 0 45 n a 1 54 1ppl 2 82 n a 5 62 lppm 0 62 n a 8 27 1pso 13 10 n a 1 61 Irbp 0 96 n a 1 13 Irds 3 75 478 4 89 Irne 10 08 2 00 12 24 irnt 0 72 n a 1 90 1rob 1 85 3 75 7 70 1slt 0 51 0 78 1 63 1snc 191 n a 7 48 1srj 0 58 042 2 36 1stp
33. not clear that a water molecule would occupy such a cavity in preference to leaving a vacuum despite the statistical terms favoring occupancy However this is a rare situation not particularly relevant to the binding of a large ligand whereas structural motifs similar to the examples in Figure 1 are quite common There have been a large number of papers in the literature studying via molecular dynamics simulations the behavior of 6180 Journal of Medicinal Chemistry 2006 Vol 49 No 21 water in contact with various types of hydrophobic structures including flat and curved surfaces 7 parallel plates 24 nanotubes and recently more realistic systems such as the hydrophobic surfaces of a protein or the interface between two protein domains 97 5 There have also been attempts to develop general theories as to how the hydrophobic effect depends on the size and shape of the hydrophobic structure presented to the water molecules 52 A number of concepts that are clearly related to the proposals in the present paper have emerged from this work evacuation of water dewetting under the appropriate conditions from regions between two predominantly hydrophobic surfaces and a model for the curvature dependence of the hydrophobic energy in which concave regions are argued to have greater hydrophobicity than convex ones However while this work provides useful ideas and general background development of a scoring function that can be
34. opening of an allosteric pocket primarily via motion of a phenylalanine residue while the second class is smaller and does not protrude into this pocket Comparing scores of these two ligand classes using a single receptor structure does not make sense If the pocket is closed the larger ligands will not fit at all whereas if the pocket is open the larger ligands will unfairly score better as the reorganization energy of the receptor required to engender the needed side chain motion will not have been included There is no overlap between the ligands associated with the two receptor forms This partitioning is meant as an introductory exploration of enrichment studies using multiple receptor conformations a topic we intend to pursue more intensively in the future Enrichment metrics to recover well docked active ligands based on number of outranking decoys and standard enrichment measures as for the training set are presented in Tables 13 and 14 respectively For all known active ligands standard enrichment measures are presented in Table 15 The same comparison database of 1000 decoy ligands employed in training set enrichment studies has been used As expected there is some quantitative degradation of the XP results from the training set but overall the results are qualitatively comparable to the training set results using the outranking decoy metric which we have argued is the most meaningful for our purposes and the improvements
35. our data set into a training set and a test set The training set screens have been used in parametrizing the XP scoring function while the test set screens have not The training set contains 15 receptors and various numbers of ligands for each receptor as enumerated in Table 7 A large and diverse training set is essential to address the range of chemical motifs identified in the XP Glide scoring function as noted previously Our present test set is relatively small and less diverse than the training set with regard to the number of new receptors considered Therefore the validation implied by the results must be considered preliminary Six receptors are considered in the test set four of which were not included in the training set vascular endothelial growth factor receptor 2 Vegfr2 peroxisome proliferators activated receptor y PPARy P secretase BACE and blood coagulation factor VIIa factor VIIa For two training set receptors CDK2 and thrombin we XP Glide Methodology and Application have located a significant number of additional ligands to include as part of the test set Training Set Results Table 7 lists for the training set the receptors investigated the number of known active ligands available with affinities better than 10 uM and the number of ligands deemed to dock correctly into the chosen receptor There is one exception active neuraminidase ligands are relatively weak binders that do not have activities bett
36. over previous versions in this case in part becausethe p38 siteis large and requires a very hydrophobic binding mode in general only one to two hydrogen bonds are made by correctly docked p38 actives Such sites represent a severe challenge for most empirical scoring functions Halgren amp al 100 18 20 25 Figure 7 Percent of actives recovered for assaying the top 296 596 and 1096 of the ranked database for the 1hpx site of HIV protease We can report however that we have made significant progress in handling sites of this naturein the ongoing development of Extra Precision Glide Detailed results for Glide 2 5 are shown in Tables S6 S8 Supporting Information 5 HIV Protease 1hpx Our screening database contains 15 ligands from cocrystallized HIV 1 protease complexes included in our docking accuracy test set Here we focus on Ihpx as thetarget The 1hpx complex retains the usual water under the flaps that endose the active site but we removed it so that ligands that displace this water such as XK263 from the 1hvr complex and A 98881 from lpro could dock The removal of this water raises the question of whether the resultant overly generous site might recognize a large number of false positives However this proved not to be the case As with the estrogen receptor screen our original docking experiments used 0 9 protein O 8 ligand scaling However this site is not especially tight and default 1 0 0
37. precision XP Glide The key novel features characterizing XP Glide scoring are 1 the application of large desolvation penalties to both ligand and protein polar and charged groups in appropriate cases and 2 the identification of specific structural motifs that provide exceptionally large contributions to enhanced binding affinity Accurate assignment of these desolvation penalties and molecular recognition motifs requires an expanded sampling methodology for optimal performance Thus XP Glide represents a single coherent approach in which the sampling algorithms and the scoring function have been optimized simultaneously The goal of the XP Glide methodology is to semiquantita tively rank the ability of candidate ligands to bind to a specified conformation of the protein receptor Because of the rigid receptor approximation utilized in Glide and other high through put docking programs ligands that exhibit significant steric clashes with the specified receptor conformation cannot be expected to achieve good scores even if they in reality bind effectively to an alternative conformation of the same receptor Such ligands may be thought of as unable to fit into that specified conformation of the protein For docking protocols to function effectively within the rigid receptor approximation some ability to deviate from the restrictions of the hard wall To whom correspondence should be addressed Phone 212 854 7606 Fax 212 854 7454
38. precomputed OPLS AA van der Waals and electrostatic grids for the receptor The energy and gradient calcula tions are performed using standard three di mensional interpolation methods The Coulomb and van der Waals fields of the protein are stored at the vertexes of a grid and the interaction of each ligand atom with these fields is evaluated using trilinear interpolation formulas for a cube Methods of this type are in common use and have been described extensively in the literature so we do not discuss the mathematical details here To ensure sufficient accuracy in regions in which the ligand and protein come into contact Glide uses a multigrid strategy The Coulomb van der Waals grid is initially built using large boxes typically 3 2 on a side and is then refined hierarchically into boxes of 1 6 0 8 or 0 4 A depending on the distance of the box to the van der Waals surface of the protein the smaller this distance the higher the resolution of the box used The resulting mesh tiles the docking volume with cubes of various size Once the identity of the cube in which a ligand atomis contained is identified the appropriately scaled interpolation formula can be used While some extra bookkeeping is required the additional computa tional cost as compared to a uniform mesh is negligible whereas the reduction in memory can be more than 2 orders of magnitude The energy minimization typically begins on a set of Coulomb and vdW grids th
39. protein ligand complexes to guide the development of a set of empirical rules outlined below for the types of ligand and 6182 Journal of Medicinal Chemistry 2006 Vol 49 No 21 Figure 3 The 1bl7 ligand interacts with p38 MAP kinase through a neutral neutral hydrogen bond between the ligand s aromatic nitrogen and the Met 109 N H group receptor chemistries that receive this type of reward These rules will likely evolve as more data is considered and further simulations are undertaken The scoring function term outlined above enables such hydrogen bonds to be detected automatically in advance of experimental measurement Of equal importance false positives which superficially share some characteristics of the required structural motif but lack a key component can be rejected automatically as well Rejection of false positives has been optimized by running a given variant of the scoring function identifying high scoring database ligands with special hydrogen bonds in locations not seen in known actives examining the resulting structure and altering the recognition function to eliminate the reward for such test cases In designing a detailed set of rules to implement the ideas outlined above we have attempted to generalize results obtained from a wide variety of ligand receptor systems while at the same time avoiding false positives and respecting the basic physical chemistry principles that form the basis of the mo
40. protein preparation is relaxation of the receptor structure so that it at least accommodates the native ligand We employ the standard Schr dinger protein preparation utility for this purpose A related issue is the use of van der Waals scaling of nonpolar ligand and protein atoms to take minor induced fit effects into account in an approximate fashion Various scalings have been examined though with the exception Journal of Medicinal Chemistry 2006 Vol 49 No 21 6191 of the human estrogen receptor 3ert which used a scaling of 0 8 on the ligand and 0 9 on the protein the standard scaling of 0 8 on the ligand and no protein scaling has been applied Only active ligands that succeed in 4 0 XP docking with the chosen scaling parameters have been retained in our enrichment studies Comparison Database Our methods for generating com parison databases are outlined in ref 1 Molecules are selected from a purchasable compound library of about one million compounds that have been filtered for predicted pharmacokinetic properties using the QikProp program Selection protocols are then applied to ensure a distribution of rings acceptors donors molecular weight and so on in line with averages determined for drug like molecules Computed Binding Affinities of Known Actives for a Wide Range of Targets As stated above our expectation is that only a small number of database ligands will be competitive with active compounds whose experi
41. required to obtain results in reasonable agreement with experiment The large number of parameters employed in turn necessitates the use of a large number of examples in the training set to avoid overfitting the training set must be substantially larger than the number of parameters that are adjusted The total number of parameters in the current XP scoring function is on the order of 80 this includes parameters for desolvation penalties hydrophobic enclosure special neutral neutral and charge charge hydrogen bonds and pi cation and pi stacking interactions Parameters are required to convert various geometrical criteria into specific scores The PDB complexes below as well as the enrichment studies in the training set were used to develop the parameter values False positives as well as known actives were incorporated into the optimization protocol so the data in the training set exceeds the total number of PDB complexes and known actives by a considerable margin although a precise calculation of the total number of data points in the training set is very difficult to produce as for example not every database ligand was competi tive with known actives in ranking and noncompetitive compounds played no role in parameter optimization Because the scoring function contains nonlinear functional forms a rigorous optimization algorithm would also be nonlinear rather than a simple least squares fit furthermore constraints would b
42. terms such effects will serve exclusively to reduce binding affinity 5 Metal Ligand Interactions Specialized terms are needed to describe the interaction of the ligand with metal ions We shall defer the discussion of metal specific parameteriza tion to another publication as this is a complex subject in its own right requiring considerable effort to treat in a robust fashion A large number of empirical scoring functions for predicting protein ligand binding affinities have been developed 5 While differing somewhat in detail these scoring functions are broadly similar A representative example the ChemScore scoring function is discussed in our comments below though similar comments would apply to many of the other scoring functions cited in refs 8 19 We briefly summarize how ChemScore treats the first four potential contributors to the binding affinity presented above 1 ChemScore contains a hydrophobic atom atom pair energy term of the form E aids pair Lp 1 ij Here i and j refer to lipophilic atoms generally carbon and f rij is a linear function of the interatomic distance rj For rj less than the sum of the atomic vdW radii plus 0 5 A fis 1 0 XP Glide Methodology and Application Between this value and the sum of atomic vdW radii plus 3 0 f ramps linearly from 1 0 to zero Beyond the sum of atomic vdW radii plus 3 0 f is assigned a value of zero This term heuristically represents the
43. the appropriately fitting protein structure is presented to the ligand and the ligand is well docked the present scoring function has a respectable ability to Friesner et al distinguish weak mM moderate uM and strong nM binders This capability is essential to the principal task in virtual screening yet is only marginally present in prior scoring functions To improve beyond this level with regard to precision we believe that receptor flexibility must be introduced and an additional level of detail with regard to the protein ligand interactions must be incorporated Robustness is another matter new moieties and chemistries seem to emerge as additional receptors are added to the test suite Thus we cannot claim to have reached convergence in this regard with our current data sets On the other hand the amount of detailed medicinal chemistry information incorporated into the current scoring function is a substantial advance as compared to alternative scoring functions in the literature A final important point is that the accuracy cited above may quantitatively degrade in cross docking calculations even when the ligand is able to assume a qualitatively correct pose in the receptor as of course would the accuracy of other docking scoring methods for similar reasons The enrichment studies presented below address this question to some extent but at present do not consider the relative rankings of different active com
44. the protein cavity which can lead to favorable entropic terms of the type discussed above in 1 Contributions to binding affinity favorable or unfavorable will also depend on the quality and type of hydrogen bonds formed net electrostatic interaction energies possibly including long range effects although these generally are considered small and typically are neglected in empirical scoring functions and specialized features of the hydrogen bonding geometry such as bidendate salt bridge formation by groups such as carboxylates or guanidium ions Finally differences in the interactions of the displaced waters as compared to the ligand groups replacing them with the protein environment proximate to the hydrogen bond can have a major effect on binding affinity as is discussed in greater detail below 3 Desolvation Effects Polar or charged groups of either the ligand or protein that formerly were exposed to solvent may become desolvated by being placed in contact with groups to which they cannot hydrogen bond effectively In contrast to the two terms described above such effects can only reduce binding affinity 4 Entropic Effects Due to the Restriction on Binding of the Motion of Flexible Protein or Ligand Groups The largest contributions are due to restriction of ligand translational orientational motion and protein and ligand torsions but modification of vibrational entropies can also contribute As in the case of desolvation
45. three 2 When a group of lipophilic ligand atoms is enclosed on two sides at a 180 degree angle by lipophilic protein atoms this type of structure contributes to the binding free energy beyond what is encoded in the atom atom pair term We refer to this situation as hydrophobic enclosure of the ligand There is some analogy here to the parallel plate nanotube with some sets of parameters and protein systems in which dewetting has been observed although the length scale of the region under consideration is smaller and likely more heterogeneous The pair hydrophobic term in eq 1 is generally fit to data from a wide range of experimental protein ligand complexes As such it represents the behavior of individual lipophilic ligand atoms in an average environment Our new terms utilize specific molecular recognition motifs and are designed to capture deviations from this average that lead to substantial increases in potency for lipophilic ligand groups of types that are typically targeted in medicinal chemistry optimization programs That is placing an appropriate hydrophobic ligand group within the specified protein region leads to substantial increases in potency Indeed the data enabling development of this term was primarily obtained from a wide range of published medicinal chemistry efforts that provided examples of lipophilic groups that yielded Friesner et al exceptional increases in potency as well as those yielding min
46. three such contacts A second more sophisticated function assembles the contacts into groups and evaluates a penalty based on the size of the contacting groups the range of contacts and the extent to which the groups lie on the periphery of the molecule Empirically it has been found that peripheral groups are more difficult to penalize for intraligand contacts than are more centrally located groups Implementation Issues Application of XP penalty terms particularly those related to desolvation imposes hurdles that make it difficult for random database ligands to achieve good scores in virtual screening These hurdles do not exist in alternative programs that ignore desolvation effects and strain energy On the other hand if the terms are inaccurately defined they will adversely affect active compounds Furthermore a definition that would be accurate for a high resolution structure may function poorly for docked structures particularly in the rigid receptor framework due to inaccuracies in sampling This issue arises routinely in practice such as when an active ligand could avoid a penalty by moving a few tenths of an Angstrom in some direction but is blocked by the rigid protein Similarly the sampling algorithm may simply fail to find the superior pose We have found that extensive sampling to enable ligands to avoid penalties when possible is an essential component of Glide XP scoring If the penalties are due to limitations
47. twice as accurate as FlexX for ligands having up to 20 rotatable bonds Glideis also found to be more accurate than the recently described Surflex method 1 Introduction The number of drug discovery projects that have a high resolution crystal structure of the receptor avail able has increased in recent years and is expected to continue to rise because of the human genome project and high throughput crystallography efforts A common computational strategy in such a case is to dock molecules from a physical or virtual database into the receptor and to use a suitable scoring function to evaluate the binding affinity A number of docking programs are employed extensively in the pharmaceuti cal and biotechnology industries 1 19 of which the most widely used appear to be GOL D FlexX and DOCK 3 Over the past several years considerable success has been reported for these programs in virtual screening applications 11 13 However none as of now can be viewed as offering a robust and accurate solution tothe docking problem even in the context of a rigid protein receptor In this paper we describe a new docking methodology that has been implemented in the FirstDiscovery soft ware package Glide grid based ligand docking with energetics Glide has been designed to perform as close to an exhaustive search of the positional orientational and conformational space available to the ligand as is To whom correspondence should be addre
48. types of hydrogen bonds The default values assigned are neutral neutral 1 0 kcal mol neutral charged 0 5 kcal mol and charged charged 0 0 kcal mol These assignments are based on a combination of physical reasoning and empirical observa tion from fitting to reported binding affinities of a wide range of PDB complexes The rationale for rewarding protein ligand hydrogen bonds at all is subtle because any such hydrogen bonds are replacing hydrogen bonds that the protein and ligand make with water At best the net number of total hydrogen bonds on average will remain the same in the bound complex as compared to solution However the liberation of waters to the bulk can be argued to result in an increase in entropy and liberation of waters around a polar protein group requires that a protein ligand hydrogen bond with similar strength be made for a desolvation penalty to be avoided This analysis is most plausible when both groups are neutral The formation of a salt bridge between protein and ligand involves very different types of hydrogen bonding from what is found in solution The thermodynamics of salt bridge formation in proteins has been studied extensively both theoretically and experimentally 2 and depends on many factors such as the degree of solvent exposure of the groups involved in the salt bridge The default value of zero that we assign is based on the presence of many protein ligand complexes in the PDB with
49. used greater scaling than the current default As we show in the section 5 the two scaling models yield comparable results Thus employing larger scaling than is needed to allow the known actives to fit into the site Enrichment Factors in Database Screening 100 100 80 80 60 60 40 40 20 20 4 18 20 25 0 18 20 25 Figure8 Percent of thrombin actives recovered for assaying the top 2 5 and 1096 of the ranked database a human thrombin 1dwc site b bovine thrombin lett site 100 100 80 80 60 60 40 40 20 20 Jm 18 20 25 18 20 25 Figure 9 Percent of actives recovered for assaying the top 296 596 and 1096 of the ranked database a thermolysin 1tmn site b Cox 2 102 site did not degrade Glide s ability to rank the known binders highly in this case Figures 2j k and 8 show that Glide 2 5 performs bet ter than either Glide 1 8 or 2 0 The same qualitative trend is seen in the EF 70 and EF 296 enrichment factors reported in Table 1 The comparisons show that the lett site consistently yields better results Tables S10 and S11 Supporting I nformation list the Glide 2 5 rankings 7 Thermolysin 1tmn As target receptor we chose the protein from the 1tmn complex The 1tmn ligand known as CLT for carboxy leu trp coordinates with the active site Zn ion via a carboxylate group and places its leucine side chain into a hydrophobic pocket Including 1tmn the docking accuracy test set contains
50. very low binding affinities in which solvent exposed protein ligand salt bridges are formed As signing the contributions of these salt bridges to the binding affinity would lead to systematically worse agreement with experimental enrichment data In XP scoring certain features of a salt bridge are required for this type of structure to contribute to binding affinity in XP scoring Finally the charged neutral default value represents an interpolation between the neutral neutral and charged charged value that appears to be consistent with the empirical data Hydrogen bond scores are diminished from their default Journal of Medicinal Chemistry 2006 Vol 49 No 21 6181 values as the geometry deviates from an ideal hydrogen bonding geometry based on both the angles between the donor and acceptor atoms and the distance The function that we use to evaluate quality is similar to that used in ChemScore In what follows specialized hydrogen bonding motifs are described in which additional increments of binding affinity are assigned in addition to those from the ChemScore like pairwise hydrogen bond term Our investigations indicate that these situations can arise for neutral neutral or charged charged hydrogen bonds but not for charged neutral hydrogen bonds The exclusion of charged neutral hydrogen bond special rewards has principally been driven by our failure to date to identify motifs of this type that help to improve the
51. 0 59 0 69 0 65 1tdb 146 10 48 10 10 Ithy 2 31 n a 2 67 Itka 228 188 1 17 1tlp 186 n a 2 85 1tmn 2 80 1 68 0 86 ltng 0 19 n a 1 93 1tnh 0 33 n a 0 56 1tni 2 18 n a 2 71 Itnj 0 35 n a 0 89 ltnk 0 87 n a 1 41 1tnl 0 23 n a 0 71 Itph 0 20 n a 1 50 Itpp 112 03 11 ltrk 1 64 n a 1 57 1tyl 106 n a 2 34 lukz 0 37 n a 0 94 lulb 028 0 32 3 37 lwap 0 12 n a 0 57 lxid 4 30 092 2 01 lxie 3 86 0 69 1 94 2ack 097 499 2 21 2ada 053 0 40 0 67 2ak3 071 5 08 091 2cgr 0 38 0 99 3 53 2cht 0 42 059 4 58 2cmd 0 65 n a 3 15 2cpp 017 n a 2 94 2ctc 161 0 322 197 2dbl 0 609 131 149 2gbp 0 15 n a 0 92 2lgs 7 55 n a 4 63 2mcp 1 30 4 37 2 07 2phh 038 072 0 43 2pk4 0 86 1 34 1 66 2plv 188 13 92 7 85 2r04 0 80 n a 12 55 2r07 0 48 823 11 63 2sim 092 0 92 1 99 2tmn 0 58 n a 5 16 2xis 0 85 n a 1 54 2yhx 3 84 119 225 2ypi 0 31 n a 1 22 3da 851 545 6 42 3cpa 240 1 58 253 3hvt 077 112 1026 3mth 5 48 10 12 1 59 3ptb 027 09 0 55 3tpi 0 49 0 80 107 4aah 0 30 042 5 93 Acts 0 19 1 57 1 53 Adfr 112 144 1 40 4fab 4 50 5 69 4 95 4fbp 0 56 n a 1 78 4fxn 0 44 n a 1 04 4hmg 0 78 n a 5 74 4phv 038 111 1 12 Atim 132 n a 4 09 4tIn 2 24 n a 3 68 4tmn 1 87 n a 8 35 4ts1 0 85 n a 1 41 5abp 0 21 n a 1 17 5cpp 0 59 n a 1 49 5cts 0 28 n a 11 61 5p2gp 18 1 55 1 00 Stim 058 n a 1 99 5tmn 2 43 n a 4 38 6abp 0 40 1 08 1 12 6cpa 4 58 n a 6 61 6rnt 2 22 1 20 4 79 6tim 1 73 n a 1 60 6tmn 2 66 n a 5 10 7cpa 4 14 n a 9 11 7tim 0 14 0 78 149 8atc 0 37 n a 0 62 8gch 030 086 8 91 9hvp 2 68 n a 15 54 Ta
52. 09 323 14 Rarey M Kramer B Lengauer T Klebe G A A Fast Flexible Docking Method Using an Incremental Construction Algorithm J Mol Biol 1996 261 470 489 15 Jones G Willet P Glen R C Leach A R Taylor R Development and Validation of a Genetic Algorithm for Flexible Docking J Mol Biol 1997 267 721 148 16 Ewing T J Makino S Skillman A G Kuntz I D DOCK 4 0 Search Strategies for Automated Molecular Docking of Flexible Molecule Databases J Comput Aided Mol Des 2001 15 411 428 2 lt 8 9 Friesner et al 17 Golke H Hendlich M Kelbe G Knowledge Based Scoring Function to Predict Protein Ligand Interactions J Mol Biol 2000 295 337 356 18 Wang R Lai L Wang S Further Development and Validation of Empirical Scoring Functions for Structure Based Binding Affinity Prediction J Comput Aided Mol Des 2002 16 11 26 19 Wang R Lu Y Wang S Comparative Evaluation of 11 Scoring Functions for Molecular Docking J Med Chem 2003 46 2287 2303 20 Wallqvist A Berne B J Computer Simulation of Hydrophobic Hydration Forces on Stacked Planes at Short Range J Phys Chem 1995 99 2893 2899 21 Lum K Chandler D Weeks J D Hydrophobicity at Small and Large Length Scales J Phys Chem B 1999 103 4570 4577 22 Nicholls A Sharp K A Honig B Protein Folding and Association
53. 13thermolysin complexes including one 1lna in which Co replaces Zn However the 2tmn 4tIn and Ihyt ligands have only about 25 atoms induding hydrogens and are too small to bind tightly for example the measured binding affinities for 2tmn and 4tln are only 5 to 7 kcal mol In this study weusethe 10 larger drug sized ligands Thermolysin is known to have a rigid active site sothe choice of target structure is probably unimportant in this case As for thrombin preliminary studies suggested a preference for using unscaled radii for both the protein and the ligand i e 1 0 1 0 scaling but the results presented here use default 1 0 0 8 scaling That unscaled radii yield good results seems dearly related tothe rigid and open nature of the site In this case somewhat better results are obtained with 1 0 1 0 scaling see section 5 but default scaling also does well As Figures 2 and 9a show Glide 2 5 performs much better than Glide 1 8 and somewhat better than Glide 2 0 For Glide 2 5 5 of the 15 top ranked ligands are known binders Table S12 Supporting I nformation One key to the improved performance is that Glide Score 2 5 spedifically rewards metal ligation by anionic ligand functionality We madethis change on the basis of experimental evidence that metalloproteases strongly J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1755 favor anionic ligands 1917 A second element is that GlideScore 2 5 considers
54. 2 44 H20 CH3S 4 0 2 4 polar polar H20 H20 3 1 3 7 function On the other hand if the ion is net neutral as it is for example in the case of the zinc metallo protein farnesyl protein transferase which accepts neutral ligands such as substituted imidazoles the preference is suppressed The seventh term from Schr dinger s active site mapping facility rewards instances in which a polar but non hydrogen bonding atom as classified by ChemScore is found in a hydrophobic region The second major component is the incorporation of contributions from the Coulomb and vdW interaction energies between the ligand and the receptor To make the gas phase Coulomb interaction energy a better pre dictor of binding and a better contributor to a composite scoring function we reduce by 50 the net ionic charge on formally charged groups such as carboxylates and guanidiniums we also reduce the vdW interaction energies for the atoms directly involved Table 1 illustrates the effect of these changes for some prototype systems The wide disparities in the original interaction energies are greatly reduced though charge charge in teractions are still favored to some extent The Coulomb vdW energies used in GlideScore 2 5 but not those used in Emodel employ these reductions in net ionic charge except in the case of anionicligand metal interactions for which Glide uses the full interaction energy The third major c
55. 2 3 5 7 10 2030 50 100 1 23 20 30 50 100 100 80 60 40 20 ma R 1 LL 20 30 50 100 1 23 710 2030 50 100 100 1 23 5710 2030 50 100 20 30 50 100 Trej 1 2 3 5710 2030 50 100 1 2 3 5 7 10 20 30 50 100 2 3 5 710 2030 50 100 Figure 2 Percent of known actives found y axis vs percent of the ranked database screened x axis for Glide 2 5 solid green Glide 2 0 blue dashed and Glide 1 8 red dot dashed Black dotted lines show results expected by chance The listed PDB codes are defined in Table 1 actives in the first 296 of the ranked database for both the mixed and pyrimidine only screens better perfor mance relative to earlier verisons of Glide is also shown in Figure 2a for the mixed screen Thereason for separating out the pyrimidines is that a labile Gln 125 side chain undergoes a 180 rotation J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1753 100 go ERR 5 100 80 60 40 20 18 2 0 25 Figure 3 a Percent of thymidine kinase 1kim actives recovered with Glide 1 8 2 0 and 2 5 for assaying the first 296 596 and 1096 of the ranked database b Percent recovered using only the seven pyrimidine based ligands as actives 18 20 25 100 100 a 80 80 60 60 40 40 20 20 18 20 25 18 20 25 Figure 4 Percent of estrogen receptor actives recover
56. 3 Figure 6 Biotin bound to streptavidin The identification of a triplet of correlated hydrogen bonds in the ring in a hydrophobically enclosed region and the three hydrogen bonds to the ligand carbonyl within that ring each contribute 3 kcal mol rewards to this tightly bound complex AGexp 18 3 kcal mol XP binding 18 2 kcal mol Figure 7 Fidarestat bound to aldose reductase The triplet of special hydrogen bonds to the ring contributes 5 0 kcal mol to the binding energy of this 9 nM inhibitor lated and experimental binding affinity using a docked structure is only 0 1 kcal mol It should be noted that the high accuracy of this prediction is fortuitous and is not intended to suggest an ability to rigorously rank order compounds Instead the intent is to contrast the qualitatively reasonable prediction with that of alternative scoring functions which typically yield results for this complex in error by 5 10 kcal mol for example as reported in ref 35 The triply correlated enclosed hydrogen bonding motif also explains the low nanomolar binding affinity in the binding of fidarestat to the lef3 structure of aldose reductase Figure 7 relative to a large number of ligands that achieve similarly large lipophilic scores in the highly hydro phobic active site yet have only micromolar affinity In our studies of various pharmaceutically relevant targets the combination of hydrophobic enclosure with one to three
57. 3 6 13 6 11 6 12 5 2 06 2 42 lbma 6 3 19 19 I 0 68 1 94 lfkg 10 9 13 1 12 6 84 1 21 1 33 Ibra 2 5 5 5 3 5 7 8 2 26 0 32 1fki 9 5 10 2 10 2 F 1 30 1 29 Ibyb 19 0 14 2 142 11 3 0 56 0 46 1fq5 11 5 17 0 17 0 14 0 1 96 2 43 Iclb 18 2 15 7 10 8 0 91 0 45 lfvt 13 2 132 84 0 88 0 88 1c3i 12 6 12 6 11 9 0 61 0 43 1g45 11 8 82 8 2 6 0 7 88 4 02 lc5p 6 4 6 2 6 2 8 6 0 27 0 25 1g46 12 1 8 2 8 2 6 3 8 06 4 50 1c83 6 6 13 13 9 9 0 17 0 14 1g48 11 5 7 0 7 0 5 9 1 88 3 77 1c84 6 8 15 19 8 8 0 26 0 32 1g4j 11 9 6 9 6 9 6 9 5 56 3 51 1c86 6 4 8 4 84 10 9 0 19 0 18 1g4o 11 3 7 6 7 6 6 0 3 39 4 21 1c87 8 1 8 1 10 9 0 28 0 21 1852 13 0 8 2 8 2 5 7 8 01 4 26 1c88 72 7 5 11 8 0 25 0 22 1g53 12 3 84 84 6 4 7 88 4 53 1c8k 8 1 8 1 i 3 28 5 50 1g54 12 0 F2 7 2 54 8 45 5 14 Icbs 9 8 1 5 7 5 cS 0 63 0 39 lghb 1 7 3 2 3 2 9 7 0 45 0 30 lcbx 8 7 7 8 7 8 13 1 0 28 0 48 lglp 4 5 4 5 8 1 0 75 0 32 Icde 15 2 15 2 11 5 1 62 1 71 lglq 26 2 6 2 9 7 0 46 0 32 lcdg 3 3 2 4 2 4 4 8 6 48 9 83 lgsp 1 8 1 10 2 79 lcil 12 9 54 54 6 2 3 61 3 92 hbv 8 7 12 1 12 1 5 5 2 19 2 07 lenx 10 0 6 8 6 8 L3 6 54 6 53 Ihde 82 9 8 9 8 8 3 0 56 0 35 lcom 5 4 7 3 ET 9 0 0 55 3 74 Ihdy 7 8 4 8 4 8 4 2 1 65 1 70 lcoy
58. 6 7 6 7 0 0 37 1 46 lida 11 9 13 1 13 1 12 2 1 95 2 12 1dwb 4 0 6 0 4 0 ERT 0 29 0 32 ligj b 92 92 7 0 0 72 0 46 1dwc 10 3 9 3 9 3 8 8 2 06 0 89 limb 5 7 6 6 6 6 10 0 1 84 1 64 ldwd 11 4 13 2 112 11 2 0 47 1 43 livb 5 2 5 2 7 0 3 27 0 47 live 34 34 5 0 2 05 1 89 lwap 8 1 8 1 10 2 0 23 0 19 livd 4 3 6 3 6 3 6 2 0 73 0 73 lxid 1 5 7 6 6 4 02 4 32 live 3 6 3 6 5 6 5 11 5 17 lxie 4 8 4 8 6 8 2 60 3 91 livf u d 6 8 0 61 0 59 2ack 9 3 9 3 7 0 1 07 0 88 llah 10 3 43 7 3 9 6 0 52 0 19 2ada 9 5 9 5 9 0 0 59 0 46 llep 9 1 SeT l 8 8 1 82 1 06 2cgr 9 9 6 5 6 5 10 8 0 56 0 52 lldm 14 6 3 6 3 1 3 1 34 1 35 2cht TS 7 7 TA 11 4 0 48 0 51 XP Glide Methodology and Application Table 5 Continued Journal of Medicinal Chemistry 2006 Vol 49 No 21 6189 GlideScore kcal mol RMS GlideScore kcal mol RMS PDB AGexp XP XP corr SP XP SP PDB AGexp XP XP corr SP XP SP llic 6 1 6 1 5 3 3 96 4 99 2cmd 6 2 8 2 82 10 3 0 65 0 34 1lmo 1 6 1 6 6 8 8 40 0 87 2cpp 8 3 8 2 8 2 6 8 0 15 3 04 llna 6 6 6 6 6 8 1 50 0 90 2ctc 5 3 7 4 7 4 10 1 1 43 1 58 list 6 1 4 1 L 0 75 0 27 2dbl 11 8 122 11 0 8 8 2 40 0 81 Imbi 2 6 3 9 3 9 4 1 1 92 1 65 2gbp 10 1 9 0 9 0 12 5 0 61 0 14 Imer 4 3 8 3 8 3 7 5 5 82 4 33 2ifb 74 8 7 6 2 2 4 2 27 1 77
59. 7 46 0 thermolysin ltmn 1 0 1 0 25 5 20 0 10 0 30 5 1 0 0 8 25 0 18 0 9 0 24 5 HIV RT Irt1 0 9 0 8 6 1 79 6 4 6 4 10 0 8 13 6 10 9 6 4 9 1 n each case the preferential scaling model used with Glide 2 0 is listed first Table 4 Number of Known Actives Docked with Negative Coulomb vdW Interaction Energies as a Function of the Protein and Ligand vdW Scale F actors for Nonpolar Atoms rank of last common no of no docked active Screen site actives 1 0 0 8 0 9 0 8 1 0 0 8 0 9 0 8 estrogen receptor 3ert 10 8 9 58 17 estrogen receptor lerr 10 9 9 87 43 HIV protease lhpx 15 15 15 408 204 Cox 2 1x2 33 21 23 490 355 HIV RT Irt1 33 30 30 632 735 Comparison to results obtained for Glide 1 8 and 2 0 shows that average measures for both early and global enrichment are 2 3 times higher for Glide 2 5 Most importantly Glide 2 5 performs significantly better for many of the more difficult screens this qualitative improvement should be borne in mind when assessing comparitive studies based on Glide 1 8 or 2 0 which have begun to appear The improved enrichment stems partly from the inclusion of scoring function terms that penalize ligand protein interactions that violate established principles of physical chemistry particularly as it concerns the exposure to solvent of charged protein and ligand groups Given reports we have received from users that earlier versions of Glide were at least competitive in database enrichme
60. 8 scaling works quite well especially for Glide 2 5 cf Figures 2i and 7 Indeed 7 ligands are found in thetop 10 ranked positions and 12 are found in the top 20 Table S9 Supporting Information This is good performance by any standard 6 Thrombin 1dwc lett Our docking accuracy test set contains five uniquethrombin inhibitors 1dwc letr Idwb 1dwd lets 1ppc and lett However only four are highly active because the 1dwb ligand benzamidine is too small to bind tightly the experi mental binding affinity is 5 4 kcal mol 1 To supple ment these four binders we induded the 12 thrombin inhibitors from Engh et al and von der Saal et al that have reported binding affinities of 10 uM or better bringing the total number of actives to 16 To preparefor the original 1dwc screen for Glide 1 8 we found that all combinations of 0 8 0 9 and 1 0 vdW scaling for nonpolar protein and ligand atoms gave strongly negative Coulomb vdW energies for the known actives and afforded reasonably negative GlideScores However the model that used unscaled vdW radii for both the protein and the ligand gave the best overall GlideScores and yielded an unfavorable hydrogen bonding score for only one ligand We therefore chose tonot scalethe vdW radii The present results however use default 1 0 protein O 8 ligand scaling Thus this screen differs in the opposite sense from the estrogen receptor and HIV protease screens which originally
61. Chem 2000 43 4759 4767 show that Glide 2 5 performs better than GOLD 1 1 FlexX 1 8 or DOCK 4 01 1 Introduction The previous paper introduced Glide a new method for rapidly docking ligands to protein sites and for estimating the binding affinities of the docked com pounds That paper described the underlying methodol ogy and showed that Glide achieves smaller root mean square rms deviations in reproducing the positions and conformations of cocrystallized ligands than have been reported for GOLD and FlexX Better docking accuracy is important in its own right in lead optimization stu dies where knowledge of the correctly docked positi on and conformation pose of a novel ligand can becrucial In lead discovery studies however docking accuracy is relevant mainly to the degree that it contributes to obtaining high enrichment in database screening We believethat accurate scoring requires accurate docking though accurate docking is not enough in itself This paper investigates the ability of Glide 2 5 run in standard precision mode to identify known binders seeded into database screens for a wide variety of pharmaceutically relevant receptors We present com parisons with earlier versions of Glide and show that very substantial progress has been made Rigorous comparisons with other virtual screening methods are difficult for us to make because we generally do not have access either to the identical sets of decoy li
62. Close in the Import Options dialog box In the Import panel select the file actorXa sp pv maegz and click Open The receptor and the ligands are imported as an entry group named factorXa sp pv The receptor is displayed in the Workspace Glide 5 5 Quick Start Guide 29 Chapter 4 Examining Glide Data 30 4 2 Viewing Poses The Project Table panel has special options for entry groups that consist of a receptor and a set of ligands These options are available from the Entry menu under View Poses when you have a single entry group selected 1 Open the Project Table panel You can do this by clicking the Open Close project table toolbar button ERE The entries in the entry group containing the receptor and the poses that you imported should be selected If not click in the Row column for the entry group 2 From the Entry menu choose View Poses Setup The receptor is fixed in the Workspace and the first pose is included in the Workspace A Mark property is added to the Project Table so you can record any poses that you want to mark as being of special interest To mark a Workspace entry type M The next few steps change the display so that the ligand fills most of the Workspace 3 From the Workspace selection toolbar button choose Molecule Rp The A on the button changes to an M to indicate that molecules are being picked 4 Click the ligand molecule The atoms in the ligand are marked with pale yell
63. Directory Before you begin the tutorial you need to create a working directory to keep all your input and output files and then make a copy of the tutorial files UNIX 1 Set the SCHRODINGER environment variable to the directory in which Maestro and Glide are installed csh tcsh setenv SCHRODINGER installation path sh bash ksh export SCHRODINGER installation path 2 Change to a directory in which you have write permission cd mydir 3 Create a directory by entering the command mkdir directory name 4 Copy the structure files to the structures subdirectory version is the 5 digit Glide ver sion number cp r SSCHRODINGER impact vversion tutorial structures This command creates the subdirectory as well as copies the files 5 In the working directory create subdirectories named glide and grids cd directory name mkdir glide grids Glide 5 5 Quick Start Guide Chapter 1 Getting Started Windows 1 Open the folder in which you want to create the folder that serves as your working direc tory The default working directory used by Maestro is your user profile which is usually set to C Documents and Settings username on XP and C NUsers username on Vista To open this folder do the following a From the Start menu choose Run b Enter USERPROFILES in the Open text box and click OK 2 Under File and Folder Tasks click Make a new folder You can also choose File Folder New 3 Enter
64. Glide related material Richard A Friesner Jay L Banks Robert B Murphy Thomas A Halgren Jasna J Klicic Daniel T Mainz Matthew P Repasky Eric H Knoll Mee Shelley Jason K Perry David E Shaw Perry Francis and Peter S Shenkin Surflex Fully Automatic Flexible Molecular Docking Using a Molecular Similarity Based Search Engine J Med Chem 2004 47 1739 1749 Thomas A Halgren Robert B Murphy Richard A Friesner Hege S Beard Leah L Frye W Thomas Pollard and Jay L Banks Glide A New Approach for Rapid Accurate Docking and Scoring 2 Enrichment Factors in Database Screening J Med Chem 2004 47 1750 1759 Richard A Friesner Robert B Murphy Matthew P Repasky Leah L Frye Jeremy R Greenwood Thomas A Halgren Paul C Sanschagrin and Daniel T Mainz Extra Precision Glide Docking and Scoring Incorporating a Model of Hydrophobic Enclosure for Protein Ligand Complexes J Med Chem 2006 49 6177 6196 Quick start guide Glide tutorial J Med Chem 2004 47 1739 1749 1739 Glide A New Approach for Rapid Accurate Docking and Scoring 1 Method and Assessment of Docking Accuracy Richard A Friesner J ay L Banks Robert B Murphy Thomas A Halgren J asna J Klicic Daniel T Mainz Matthew P Repasky t Eric H Knoll t Mee Shelley J ason K Perry David E Shaw Perry Francis and Peter S Shenkin Department of Chemistry Columbia University New Yo
65. If you did select this option the Write pose viewer file option is automatically selected 12 Start the job with the name factorXa xp refine This job may take up to 30 minutes to run on a 2 GHz processor Glide 5 5 Quick Start Guide 27 28 Glide 5 5 Quick Start Guide Chapter 4 Examining Glide Data In this chapter Glide results are examined with an emphasis on visual rather than numerical appraisal The first set of exercises use the Project Table to display the results of a Glide docking job examine individual ligand poses and their contacts with the input receptor struc ture The second set of exercises uses the Glide XP Visualizer panel to display information on the terms in the Glide XP scoring function that contribute to the ligand binding If you have not started Maestro start it now see Section 1 3 Before proceeding with the exercises change the working directory to the glide directory See Section 1 3 on page 3 for instructions on how to do this 4 1 Importing Pose Data The first task is to import poses from a pose viewer file into the Maestro project iR 5 6 In the main window on the toolbar click the Import structures button E The Import panel opens Click on Options The Import Options dialog box opens Ensure that Maestro is chosen from the Files of type menu Ensure that Import all structures Replace Workspace and Fit to screen following import are all selected Click
66. Insights from the Interfacial and Thermodynamic Properties of Hydrocarbons Proteins 1991 11 281 296 23 Zhou R H Huang X H Margulis C J Berne B J Hydrophobic collapse in multidomain protein folding Science 2004 305 1605 1609 Huang X H Zhou R H Berne B J Drying and hydrophobic collapse of paraffin plates J Phys Chem B 2005 109 3546 3552 24 Cheng Y K Rossky P J Surface Topography Dependence of Biomolecular Hydrophobic Hydration Nature 1998 392 696 699 25 Hummer G Rasaiah J C Noworyta J P Water Conduction Through the Hydrophobic Channel of a Carbon Nanotube Nature 2001 4 4 188 190 26 Liu P Huang X H Zhou R H Berne B J Observation of a dewetting transition in the collapse of the melittin tetramer Nature 2005 437 159 162 27 Wallqvist A Berne B J Molecular Dynamics Study of the Dependence of Water Solvation Free Energy on Solute Curvature and Surface Area J Phys Chem 1995 99 2885 2892 28 Lee S H Rossky P J A Comparison of the Structure and Dynamics of Liquid Water at Hydrophobic and Hydrophilic Surfaces A Molecular Dynamics Simulation Study J Chem Phys 1994 100 3334 3345 Lee C Y McCammon J A Rossky P J The Structure of Liquid Water at an Extended Hydrophobic Surface J Chem Phys 1984 80 4448 4455 29 Sharp K A Nicholls A Fine R F Honig B Reconciling the Magnitude of the Microscop
67. OLD retains waters for 2ctc 1mdr and 1nis and FlexX does so for 1aaq 1lna 1xie and 4phv Omitting structural waters can be useful because it may allow ligands to be found that are capable of displacing them For example we removed the water under the flaps in HIV protease to allow ligands such as the DuPont Merck cyclic urea which displaces this water to dock Theresultant overly generous active site might have encouraged false positives in the docking H owever excellent rank orders were obtained for known HIV ligands in the database screen 16 The fifth step adds hydrogens to the protein to any cofactors and to any added structural waters and the final step carries out a series of restrained minimiza tions on the protein ligand complex The first stage reorients repositionable side chain hydroxyls in Ser Thr and Tyr residues and side chain sulfhydryls of Cys This is accomplished by tightly tethering non hydrogen atoms force constant 10 kcal mol and mini mizing the hydrogens with torsion interactions turned off In effect the hydrogens are allowed to fly freely in the electrostatic wind Subsequent steps restore the torsion potential and use progressively weaker re straints on the non hydrogen atoms hydrogen atoms are always free the force constants employed are 3 1 0 3 and 0 1 kcal mol 2 The procedure stops when the cumulative rms deviation from the initial coordi nates for non hydrogen atoms
68. Table 1 show that the calculated enrichment factors are relatively low when based on all 33 Cox 2 ligands When recomputed to count only the 23 site 1 ligands as actives however Glide 2 5 yields a quite decent EF 70 value of 14 2 Thus Glide 2 5 is effective at finding active Cox 2 ligands it just cannot 1756 J ournal of Medicinal Chemistry 2004 Vol 47 No 7 100 100 a 80 80 AIL RW 4 T 40 40 ii i i 18 20 25 18 20 25 Figure 10 Percent of HIV reverse transcriptase actives recovered for assaying the top 296 596 and 1096 of the ranked database a Irt1 site b 1vrt site dock and score all of them well when the 1cx2 site is used 9 HIV RT 1vrt 1rtl For HIV reverse tran scri ptase we docked a set of 33 active ligands into the non nudeoside binding site used by nevirapine Sustiva and other NNRTI compounds The active ligands taken from a variety of literature sources include nevirapine and MKC 442 We used both the nevirapine site 1vrt and the MK C 442 site 1rt1 as targets On the basis of the dockings of the active ligands we chose 0 9 protein 0 8 ligand scaling for 1rt1 and 1 0 protein O 8 ligand scaling for 1vrt when testing Glide 2 0 The present results however use default 1 0 0 8 scaling for both sites Table 1 and Figures 2n o and 10 show that Glide 2 5 greatly outperforms the earlier releases for these two HIV RT sites This site is also very hydrophobic and offers few hydrogen bonding o
69. ae gz and click Open Ensure that the selected Range is from 1 to End the default Start the job with the name factorXa sp core The job takes a similar time to the constraints job When the job finishes examine the results in the Project Table 3 9 Refining Docked Ligands with Glide XP In this exercise you will use Glide XP to refine a set of ligands taken from the first SP docking run 1 In the Core tab select Do not use Core constraints are now turned off Glide 5 5 Quick Start Guide Chapter 3 Ligand Docking 2 In the Settings tab under Precision select XP extra precision 3 Under Options select Refine do not dock 4 Select Write XP descriptor information Note This option requires a special license If you do not have this license do not select the option You will obtain results without the license but you will not be able to complete the exercise in Section 4 5 on page 33 5 In the Ligands tab ensure that File is selected 6 Click Browse A file selector is displayed 7 Ensure that Files of type is set to Maestro 8 Navigate to the tutorial structures directory 9 Choose refine xp entries mae gz and click Open 10 Ensure that the selected Range is from 1 to End the default 11 If you did not select Write XP descriptor information in the Structure output section of the Output tab select Write pose viewer file includes receptor filename will be lt job name pv mae gz
70. ands can score ahead of underpredicted active compounds The crucial goal however is not to achieve perfection but rather to eliminate systematic errors that can lead to large numbers of false positives and little enrichment Neglect of special neutral neutral hydrogen bond terms such as those discussed in section 2 is an example of a systematic error of this nature One would predict that any method that neglects this term should exhibit poor enrichment factors for kinases such as CDK2 where such hydrogen bonds are a crucial component of the molecular recognition motif This is because we have found that large hydrophobic compounds with a few strategically placed polar groups can fit into the active site and form structures that based on the usual pair scoring terms are highly competitive with the known actives Another and perhaps the ultimate measure of performance of the scoring function is whether high ranking database ligands are in fact active The best way to address this issue is via experimental testing of such database ligands We are in the process of carrying out such tests Training Set Composition Our training set primarily focused on pharmaceutical targets of current interest is pre sented in Table 7 This suite represents a wide variety of 6192 Journal of Medicinal Chemistry 2006 Vol 49 No 21 Friesner et al Table 8 Average Attractive Components of 4 0 XP Score from Eq 3 of Correctly Docked Active Li
71. angle is close to 90 When the angle between lipophilic protein atoms is close to 180 we have argued this leads to an especially poor environment for waters 4 Each lipophilic ligand atom is assigned a score based on the number of total lipophilic contacts with protein atoms weighted by the angle term If no protein atom is greater than 90 degrees from the anchor atom the angle term is zero and the atom contributes zero to the group s Enya_enclosure term The overall score for a group is the sum over all atoms in that group of the product of the angular factor and a distance dependent factor 5 If the score for any ligand group is greater than 4 5 kcal mol the penalty is capped at 4 5 kcal mol This was an empirical determination based on investigating many test cases and comparing the results with experimental data The capping is rationalized by arguing that if a very large region of this type leads to a score greater than 4 5 kcal mol there is probably some ability of the water molecules to compensate by interacting with each other An experimentally validated example of the gain in binding affinity from placing a large hydrophobic group in a pocket in which lipophilic protein atoms are present on both sides of the pocket rings in both cases is shown in Figure 2 Here replacing a phenyl substituent with a naphthyl group was shown to result in a 21 fold improvement in experimentally measured affinity Ka The naphthyl is requ
72. ar conformation is chosen from the Amide bonds option menu These are the defaults The receptor grids and the basic Glide settings for the ligand docking job are now specified In the next section you will specify a set of ligands to dock and in the following section you will specify output options The options in the remaining three tabs Core Constraints and Simi larity can be left at their defaults for this exercise Glide 5 5 Quick Start Guide Chapter 3 Ligand Docking Ligand Docking lolx Settings Ligands Core L Constraints i Similarity Output Ligands to be docked We strongly recommend that you prepare the ligands before docking for example with LigPrep or MacroModel Use ligands from File File name all glide 2008u1 st ructures 50ligs mae g Range fa to 166 X End O Selected entries Workspace _ Use input partial charges Do not dock or score ligands with more than 300 atoms Do not dock or score ligands with more than 50 rotatable bonds Scaling of van der Waals radii To soften the potential for nonpolar parts of the ligand you can scale the vdW radii of ligand atoms with partial atomic charge absolute value less than the specified cutoff No other atoms in the ligand will be scaled Scaling factor 0 8 Partial charge cutoff 0 15 Start Write Reset Close Help
73. at have been smoothed to reduce the large energy and gradient terms that result from too dose interatomic contacts it finishes on the full scale OPLS AA nonbonded energy surface anneal ing This minimization consists only of rigid body translations and rotations when external conformations are docked rigidly For flexible docking however the minimization also includes torsional motion about the core and end group rotatable bonds Finally the three to six lowest energy poses obtained in this fashion are subjected to the Monte Carlo procedure cited in sec tion 2 The minimized poses are then rescored using the scoring function described in section 3 As previously noted the choice of the best docked structure is made using a model energy score E model that combines the energy grid score the binding affinity predicted by Method and Assessment of Docking Accuracy GlideScore and for flexible docking the internal strain energy for the model potential used to direct the conformational search algorithm Optimization of the Scoring Function The pa rameter optimization for GlideScore 2 5 used a simulated annealing algorithm that has proven to be very efficient at producing large changes to the input values for the parameters when large changes are warranted In addition to the terms described in section 3 the fitting process also considered but ultimately rejected a number of other prospective terms including nearly all
74. at lie within nominal limits and a partial score 1 00 0 00 for distances or angles that lie outside those limits but insidelarger threshold values For example g Ar is 1 00 if the H X hydrogen bond distance is within 0 25 of a nominal value of 1 85 but tails off tozeroin a linear fashion if the distance lies between 2 10 and 2 50 Similarly h Aa is 1 00 if the Z H X angle is within 30 of 180 and decreases to zero between 150 and 120 Method and Assessment of Docking Accuracy Glide 2 5 employs two forms of GlideScore i Glide Score 2 5 SP used by Standard Precision Glide ii GlideScore 2 5 XP used by Extra Precision Glide These functions use similar terms but are formulated with different objectives in mind Specifically GlideScore 2 5 SP is a softer more forgiving function that is adept at identifying ligands that have a reasonable propensity to bind even in cases in which the Glide pose has significant imperfections This version seeks to minimize false negatives and is appropriate for many database screening applications In contrast GlideScore 2 5 XP is a harder function that exacts severe penalties for poses that violate established physical chemistry prin ciples such as that charged and strongly polar groups be adequately exposed to solvent This version of Glide Score is more adept at minimizing false positives and can be especially useful in lead optimization or other studies in which only a limi
75. ate should be more favorable than binding a single ion An example of a zwitterion binding in this fashion is shown in Figure 9 XP Glide Methodology and Application Table 3 Electrostatic Rewards Note that Double Counting Is Avoided reward charge interaction kcal mol charged ligand atom in low electrostatic 1 5 potential environments zwitterion configuration range of rewards 3 0 to 4 7 increasing with electrostatic attraction positive ligand group binding to weakly 0 5 solvated negative protein group ligand CO2 group hydrogen bound to multiple 1 0 proximate positive protein residues salt bridge pair in low solvation environment 2 0 less than nine second shell waters about pre ligated charge protein atom Table 4 4 0 XP Binding Energies for Docking into Various GIuR2 Receptors Compared to Experiment AGexp ligand XP score kcal mol 1ftl 8 9 83 lpwr 12 6 13 0 1ftj 6 6 8 5 1mm7 10 7 127 Imgi 11 9 11 9 lftm 11 3 8 8 1nOt 6 6 5 7 1m5b 9 3 9 3 1m5c 9 4 9 3 1m5e 6 4 6 3 a These systems were used to calibrate the charged charged hydrogen bond recognition motif Enb_cc_motit 4 Cases where the ligand is positively charged and the protein is negatively charged are distinguished from those in which the charge states are reversed 5 Strength of the electrostatic field at the ligand An enhanced binding affinity for a salt bridge is assigned if the site at
76. auer T Klebe G A A fast flexible docking method using an incremental construction algorithm Chem Biol 1996 261 470 489 Ewing T J A Kuntz I D Critical evaluation of search algorithms for automated molecular docking and database screening J Comput Chem 1997 18 1175 1189 Ewing T J J A Makino S Skillman G Kuntz I D DOCK 4 0 search strategies for automated molecular docking of flexible molecule databases J Comput Aided Mol Des 2001 15 411 428 Abagyan R Totrov M Kuznetsov D ICM a new method for protein modeling and design Application to docking and struc ture prediction from the distorted native conformation J Comput Chem 1994 15 488 506 Abagyan R Totrov M High throughput docking for lead generation Curr Opin Chem Biol 2001 5 375 382 5 Welch W Ruppert J J ain A N Hammerhead fast fully automated docking of ligands to protein binding sites Chem Biol 1996 3 449 462 McMartin C Bohacek R QXP powerful rapid computer algorithms for structure based drug design J Comput Ai ded Mol Des 1997 11 333 344 Morris G M Goodsell D S Halliday R S Huey R Hart W E Belew R K Olson A Automated docking using a Lamarkian genetic algorithm and an empirical binding free energy function J Comput Chem 1998 19 1639 1662 Murray C W Baxter C A Frenkel A D The sensitivity of the results of mo
77. b is complete the status is changed to completed finished The opening of the Monitor panel depends on a pref erence that is set in the Jobs tab of the Preferences panel The job takes approximately 10 minutes on a 1 GHz Pentium 4 processor this time may vary depending on your particular system configuration and workload Before the job is launched these job input files are written factorXa grid in Command input for grids job factorXa grid maegz Receptor structure input for grids job When the calculation is complete the grids directory will contain the following output files factorXa grid log Log summary file from grids job factorXa grid out Output summary file from grids job factorXa grid zip Archive containing grid files The zip archive contains the following grid files FactorXa grid csc factorXa grid gsc FactorXa grid site factorXa grid save FactorXa grid greedy save factorXa grid cons FactorXa grid recep mae factorXa grid grd FactorXa grid coul2 fld factorXa grid vdw fld FactorXa grid vdwc In the next chapter you will dock ligands using these grids You can close the Receptor Grid Generation panel now Glide 5 5 Quick Start Guide Chapter 3 Ligand Docking The exercises in this chapter demonstrate the use of Glide to screen a multiple ligand file for structures that interact favorably with a receptor active site The receptor grid files you calcu lated in the previous chapter will be used to
78. ble 6 Accuracy in Cross Docking of Thymidine Kinase Inhibitors to the 1kim Site Table 7 Effect of Input Ligand Geometry on Docking Accuracy for Glide Using the GOLD Test Set rmsd of best scoring pose ligand Glide DOCK Flexx GOLD Surflex dT 0 45 0 82 0 78 0 72 0 74 ahiu 0 54 1 16 0 88 0 63 0 87 met 0 79 7 56 1 11 1 19 0 87 dhbt 0 68 2 02 3 65 0 93 0 96 idu 0 35 9 33 1 03 0 77 1 05 hmtt 2 83 9 62 13 30 2 33 1 78 hpt 1 58 1 02 4 18 0 49 1 90 acv 4 22 3 08 2 71 2 74 3 51 gcv 3 19 3 01 6 07 3 11 3 54 pcv 3 53 4 10 5 96 3 01 3 84 a The last three ligands are purines that are not expected to fit properly into a pyrimidine based site such as 1kim see text Data for DOCK FlexX and GOLD are taken from Rognan and co workers data for Surflex are taken from J ain 30 quite reasonable results in this case but FlexX and DOCK fare noticeably more poorly Influence of Input Ligand Geometry on Docking Accuracy Table 7 examines the effect on the final ligand set av rms native ligands 1 49 M MFF 94s optimized native ligands 2 00 M MFF 94s conformational search 1 85 Corina 2 35 M MFF 94s optimized Corina 2 02 docked structures of using different starting ligand geometries The native ligands perform best followed by the conformationally optimized ligands The M MFF 94s optimized geometries used for the native ligands and the Corina structures do slightly less well but are docked more accurately than ar
79. bound to tern N9 influenza virus neuraminidase combination of restricted water access for the protein group and an exceptionally strong electrostatic interaction between the ligand and protein group that creates the molecular recognition motif 3 Zwitterion ligands A principal reason that the default value for charged charged hydrogen bonds is set at zero is that in forming a salt bridge both the protein and ligand must surrender long range contributions to the Born energy i e those beyond the first shell Satisfying the first shell complement of hydrogen bonds is quite possible when forming a salt bridge but the replacement of bulk water with the protein or bound waters clearly reduces the possible dielectric response to the ion For a monovalent ion the unscreened Coulomb field decreases as 1 r Even though past the second shell dielectric screening substantially reduces further contributions long range effects make a nontrivial contribution to the total solvation free energy However for a zwitterion the fields from the positive and negative charges to some extent cancel at long range yielding a dipolar field for which the second and higher shell contributions to the solvation free energy are significantly reduced This cancellation depends on the separation of the two charged groups Thus formation of two salt bridges by the zwitterion particularly if the two oppositely charged moieties in the ligand are spatially proxim
80. can be made to Glide with regard to both accuracy and computational efficiency The algorithms for initial screening and energy mini mization are not yet fully optimized and a 2 to 3 fold reduction in computational effort per ligand may be attainable The ability to impose constraints on the ligand position e g by requiring that a suitable ligand atom be hydrogen bonded to a particular protein residue or be coordinated to a metal atom in the protein is included in Glide 2 5 This approach allows the user to guarantee satisfaction of targeted protein interactions and speeds up the calculations by reducing the size of the phase space that needs to be examined We expect to continue to improve the scoring functions used in all three phases of the calculation initial screening energy minimization binding affinity prediction Most impor tantly we have developed new sampling and scoring algorithms for Extra Precision Glide that are proving to be highly efficient at finding correctly docked poses and at rejecting false positives in database screens these efforts will be described in a subsequent paper 6 Methods Conformation Generation As a first step in its docking protocol Glide carries out an exhaustive con formational search augmented by a heuristic screen that rapidly eliminates conformations deemed not to be suitable for binding to a receptor such as conformations that have long range internal hydrogen bonds This procedure
81. chment studies as if not better than alternatives such as GOLD and FlexX One would therefore expect 4 0 XP to outperform these methods by a margin similar to that seen in Table 9 for 4 0 SP Results of Test Set Enrichment Studies A summary of key data for our test set including the receptor crystal structures used and the number of known and well docked actives again restricted to ligands with experimental binding affinities better than 10 uM are presented in Table 12 The test set includes two kinases CDK2 Vegfr2 three proteases thrombin BACE factor VIIa and one nuclear hormone receptor PPAR and hence is reasonably diverse with regard to function all of the receptors in the test set are drug targets of current or recent interest There is less diversity with regard to active site size and hydrophobicity than in the training set As discussed above validation with a larger test set will be addressed in future publications All test set calculations were performed with the released versions of Glide 4 0 XP and SP with no parameter adjustment being made to improve results for any targets For two of the receptors Vegfr2 and PPARy we utilize two different forms of the receptor structures These are highly flexible active sites and a significant fraction of ligands in both cases can be divided into groups that clearly fit better into one version of the receptor or the other For PPARy for example one class of ligands requires
82. correctly positioned hydrogen bonds is characteristic of every special neutral neutral hydrogen bond motif that leads to an exceptional increase in experimentally measured potency However additional characteristics are required to eliminate false positives In particular if the hydrogen bond partner in the protein is highly solvent exposed formation of a structure capable of solvating the group in question while still allowing the waters involved to form a suitable number of additional 6184 Journal of Medicinal Chemistry 2006 Vol 49 No 21 Table 2 Special Hydrogen Bond Reward Values reward hydrogen bond moiety kcal mol single bond to ring in hydrophobic environments L5 neutral pair in hydrophobic environments 3 0 hydrogen bonds becomes easier Thus we require that the protein group s involved in the special hydrogen bond s have a limited number of waters in the first or second shell Determination of the number of surrounding waters is carried out via the water addition code described later in this section The magnitude of the rewards associated with the special hydrogen bonds has been determined by optimization against a large experimental database containing a significant number of examples of each type of structure Values are given in Table 2 It is conceivable that finer discriminations depending upon the details of the donors and acceptors hydrophobic environ ment bound waters and so on could be d
83. cussion and Conclusions This paper has presented results for 15 database screens covering 9 widely varying receptor types Using recovery of 7096 of the known actives as a benchmark Glide 2 5 yields enrichment factors of at least 10 for all but CDK 2 p38 Cox 2 and HIV RT and of less than 5 for only the 1a9u 1bl7 and 1kv2 sites for p38 For Cox 2 the modest EF 60 value found when all 33 actives are considered is mitigated by the finding that Glide 2 5 places 9 of the 33 known binders in the first 20 ranked positions HIV RT has long been problematic but Glide 2 5 treats it considerably better than did its predecessors Two of the most troublesome remaining Screens are p38 and CDK 2 but progress is observed here too when XP Glide is employed 13 1758 J ournal of Medicinal Chemistry 2004 Vol 47 No 7 Halgren amp al Table 3 Sensitivity of Calculated Enrichment F actors to vdW Scaling Sactors vdW scaling enrichment factor screen site protein ligand EF 290 EF 590 EF 1096 EF 70 thymidine kinase tk lkim 1 0 0 9 20 0 12 0 9 0 17 9 1 0 0 8 25 0 12 0 9 0 17 9 tk pyrimidine 1 0 0 9 21 4 14 3 8 6 21 9 1 0 0 8 28 6 14 3 8 6 22 5 estrogen receptor 3ert 0 9 0 8 40 0 16 0 8 0 70 7 1 0 0 8 35 0 14 0 8 0 75 0 estrogen receptor lerr 0 9 0 8 35 0 18 0 9 0 60 4 1 0 0 8 30 0 14 0 9 0 41 2 thrombin 1dwc 1 0 1 0 12 5 10 0 6 9 10 3 1 0 0 8 15 6 11 2 7 5 11 6 HIV protease Ihpx 0 9 0 8 36 7 18 7 9 3 40 1 1 0 0 8 40 0 17 3 8
84. d Docking Job 20 3 6 Docking in High Throughput Virtual Screening Mode 21 3 7 Docking Ligands Using Constraints see 22 3 8 Docking Ligands Using Core Constraints esses 24 3 9 Refining Docked Ligands with Glide XP s sse 26 Glide 5 5 Quick Start Guide Contents Chapter 4 Examining Glide Data na an an 29 4 1 Importing POSE Data ee 29 4 2 VIGWING POS6S 22u n uu2 s seele 30 4 3 Displaying Atoms by Proximity ese 31 4 4 Visualizing Hydrogen Bonds and Contacts sss 32 4 5 Visualizing Glide XP Descriptors sse 33 4 6 Finishing the Exercises eerte tea ere nahen 34 Geting Fr Deere 35 BDA ee 39 Glide 5 5 Quick Start Guide Document Conventions In addition to the use of italics for names of documents the font conventions that are used in this document are summarized in the table below Font Example Use Sans serif Project Table Names of GUI features such as panels menus menu items buttons and labels Monospace SSCHRODINGER maestro File names directory names commands envi ronment variables and screen output Italic filename Text that the user must replace with a value Sans serif CTRL H Keyboard keys uppercase Links to other locations in the current document or to other PDF documents are colored like t
85. d proteins We focus here however on the case in which an experimental structure is available Our procedure normally starts with a protein and a cocrystallized ligand It finishes with a partially opti mized protein ligand complex to which hydrogens have been added subject to adjustment of protonation states for ionizable residues modification of tautomeric forms for histidine residues and repositioning of reorientable hydrogens e g side chain hydroxyl hydrogens The first step is to prepare the cocrystallized ligand by making sure that multiple bonds are defined cor rectly and that hydrogens are properly added Normally proteins are provided without attached hydrogens When hydrogens are present we usually delete all except those in peptide bonds In such a case it is important to note where the authors of the structure have assigned nonstandard protonation states so that these can be reimposed later if this appears warranted Cofactors which are included as part of the protein need to have multiple bonds and formal charges as signed properly so that hydrogens will be added cor rectly in a later step The second step uses the pprep script that is provided with Glide This procedure neutralizes residues that do not participate in salt bridges and that are more than a specified distance from the nearest ligand atom By default it choses the value between 10 and 20 for this distance that minimizes the total charge of the
86. de fl exx html flexx eval htm l 30 J ain A N Surflex fully automatic flexible molecular docking using a molecular similarity based search engine J Med Chem 2003 46 499 511 31 CMC database is available from MDL Information Systems Inc San Leanrdo CA 32 Oprea T Property distribution of drug related chemical databases J Comput Aided Mol Des 1999 14 251 264 See Figure 2 in this reference 33 This distance is defined as the distance from a heteroatom that bears or shares a positive or negative formal charge in the ionized protein residue to the nearest ligand atom 34 Impact is the computational engine of Schr dinger s FirstDis covery suite I mpact was developed in the laboratories of Prof Ronald Levy Rutgers University It and the FirstDiscovery suite are available from Schr dinger L L C New York 35 MacroM odel formerly known as BatchMin is a general purpose molecular mechanics program available from Schr dinger L L C New York MacroM odel was developed in thelaboratories of Prof Clark Still Columbia University J M0306430 1750 J Med Chem 2004 47 1750 1759 Glide A New Approach for Rapid Accurate Docking and Scoring 2 Enrichment Factors in Database Screening Thomas A Halgren Robert B Murphy t Richard A Friesner Hege S Beard Leah L Frye W Thomas Pollard and J ay L Banks Schr dinger L L C 120 W 45th Street New York New York 10036 Depar
87. del The detailed algorithm for detecting a single hydrogen bond in a hydrophobic environment is outlined as follows In our implementation the donor or acceptor atom must be in a ring with the exception of nitrogen which is allowed to be a nonring atom If the ligand atom is in a donor group then all other donors of the group e g the two hydrogen atoms of NH2 must be hydrogen bonded to the protein Only backbone protein atoms can participate in this type of special hydrogen bond The unligated protein donor or acceptor must have fewer than three first shell solvating waters where the waters are placed as outlined in section 3 If these criteria are met the sum of the angular factor of the hydrophobic enclosure packing score described in this section is made over the hydrogen bonded ligand heavy atoms and the carbon atoms attached to this ligand atom The hydrophobic enclosure packing score used in this sum contains only the angular weight of the score described earlier and not the distance based weight If the absolute hydrophobic enclosure packing sum is above a defined cutoff the hydrogen bond is considered to be in a hydrophobically constrained environment and a special hydrogen bond reward of 1 5 kcal mol is applied to this hydrogen bond In some instances it is found that a small perturbation of the ligand can move the hydrophobic sum above the cutoff Therefore if the hydrophobic sum is below the cutoff small rigid body perturba t
88. demonstrated Finally in the conclusion we summarize our results and discuss future directions 2 Glide XP Scoring Function The major potential contributors to protein ligand binding affinity can readily be enumerated as follows 1 Displacement of Waters by the Ligand from Hydro phobic Regions of the Protein Active Site Displacement of these waters into the bulk by a suitably designed ligand group will lower the overall free energy of the system Waters in such regions may not be able to make the full complement of Friesner et al hydrogen bonds that would be available in solution There are also entropic considerations if a water molecule is restricted in mobility in the protein cavity release into solvent via ligand induced displacement will result in an entropy gain As one ligand releases many water molecules this term will contribute favorably to the free energy Replacement of a water molecule by a hydrophobic group of the ligand retains favorable van der Waals interactions while eliminating issues concerning the availability of hydrogen bonds Transfer of a hydrophobic moiety on the ligand from solvent exposure to a hydrophobic pocket can also contribute favorably to binding by withdrawing said hydrophobic group from the bulk solution 2 Protein Ligand Hydrogen Bonding Interactions as well as Other Strong Electrostatic Interactions Such as Salt Bridges In making these interactions the ligand displaces waters in
89. depth of 0 1 to 0 2 kcal mol FlexX does not use a molecular mechanics expression for nonbonded repulsion I ndeed it may use no repulsive function at all though it does reject dockings for which the overlap volume exceeds 2 5 for any particular pair of ligand and protein atoms or 1 0 averaged over all such interactions 28 In view of these differences it is not clear that the results for GOLD and FlexX would be materially improved by using our preparations H owever only explicit compari sons that use identical protein preparations can resolve this question Comparison to Surflex Docking accuracy is also better than that recently reported by J ain for Surflex 30 In particular for a common set of 78 cocrystallized PDB complexes of 81 considered by J ain Glide gives an average rmsd from the cocrystallized ligand pose of 1 35 whereas Surflex gives an average rmsd of 1 82 Moreover Glide places 47 of the 78 ligands wthin 1A rmsd as against 38 for Surflex and makes errors of more than 2 in 14 cases as against 19 for Surflex Cross Docking for Thymidine Kinase Table 6 shows rms deviations to the cocrystallized pose for docking of thymidine kinase actives to the 1kim site by GOLD FlexX and DOCK by Surflex 2 and by Glide All methods have trouble docking the acv gcv and pcv ligands This is expected because acv gcv and pcv are purine based ligands that do not fit properly into the pyrimidine based 1kim si
90. der Waals radii of receptor atoms with partial atomic charge absolute value less than the specified cutoff All other atoms in the receptor will not be scaled Scaling factor Partial charge cutoff 0 25 Per atom van der Waals radius and charge scaling Per atom scale factors None Read from input structure file must be Maestro format Specify for selected residues VdW radius scale factor 1 0 Charge scale factor 1 0 Select residues for scaling factor Pick Residues v Selection Previous Select I I Delete Delete All _ Use input partial charges Start Write Reset Close Help Figure 2 1 The Receptor Grid Generation panel 2 3 Defining the Active Site Now that the ligand has been excluded the volume for which grids will be calculated can be defined 1 Click the Site tab The entire complex is shown with several types of markers the dark green ligand mole cule markers that appeared when the ligand was identified and the new markers that appeared when the Site tab was opened The enclosing box is shown in purple The center of the enclosing box is marked by green coordinate axes Glide 5 5 Quick Start Guide Chapter 2 Receptor Grid Generation The purple enclosing box represents the volume of the protein for which gr
91. displacement of waters from hydrophobic regions by lipophilic ligand atoms Numerous close contacts between the lipophilic ligand and protein atoms indicate that poorly solvated waters have been displaced by lipophilic atoms of the ligand that themselves were previously exposed to water The resulting segregation of lipophilic atoms and concomitant release of waters from the active site lowers the free energy via the hydrophobic effect which is ap proximately captured by the pair scoring function above Terms based on contact of the hydrophobic surface area of the protein and ligand while differing in details essentially measure the same free energy change and have a similar physical and mathematical basis Various parameterizations of the atom atom pair term have been attempted including efforts such as PLP in which every pair of atom types is assigned a different empirical pair potential However it is unclear whether this more detailed parameteriza tion yields increased accuracy in predicting binding affinities A key issue is whether a correct description of the hydrophobic effect can be achieved in all cases by using a linearly additive pairwise decomposable functional form 2 ChemScore evaluates protein ligand hydrogen bond quality based on geometric criteria but otherwise does not distinguish between different types of hydrogen bonds or among the differing protein environments in which those hydrogen bonds are embedded
92. dock ligands from the file 501igs mae Most of the 50 ligands in the file are decoys selected as a representative sample from a database of druglike molecules using the 1igparse utility Four ligands out of the total of 50 are active ligands for the chosen receptor These ligands have all been prepared for docking with LigPrep see Section 3 4 of the Glide User Manual or the LigPrep User Manual for more information on ligand preparation Typically Glide standard precision docking is used to find probable good binders in a large set the top scoring 10 to 30 can then be investigated more intensively using Glide extra preci sion XP docking or other methods available from Schr dinger In these exercises you will use all three docking modes HTVS SP and XP and also investigate the use of constraints If you have not started Maestro start it now see Section 1 3 Before proceeding with the exercises change the working directory to the glide directory See Section 2 1 on page 5 for instructions on how to do this 3 1 Specifying a Set of Grid Files and Basic Options In this exercise you will select the grid that you calculated in Chapter 2 for the ligand docking job and set the basic docking options 1 Click the Clear workspace toolbar button P 2 From the Applications menu choose Glide Ligand Docking The Ligand Docking panel opens with the Settings tab displayed 3 In the Receptor grid section click the Browse button
93. docking and scoring 1 Method and assessment of docking accuracy J Med Chem 2004 47 1739 1749 2 Schr dinger L L C New York 3 J ones G Wilett P Glem R C Leach A R Taylor R Development and validation of a generic algorithm and an empirical binding free energy function J Mol Biol 1997 267 727 748 4 Rarey M Kramer B Lengauer T Klebe G A A fast flexible docking method using an incremental construction algorithm Chem Biol 1996 261 470 489 5 Rognan D Bioinformatic Group UMR CNRS France Personal communication to T A H Bissantz C Folkers G Rognan D Protein based virtual screening of chemical databases 1 Evaluation of different dock ing scoring combinations J Med Chem 2000 43 4759 4767 Engh R A Brandstetter H Sucher G Eichinger A Bauman U Bode W Huber R Poll T Rudolph R van der Saal W Enzymeflexibility solvent and weak interactions characterize thrombin ligand interactions implications for drug design Structure 1996 4 1353 1362 8 von der Saal W Kucznierz R Leinart H Engh R A Derivatives of 4 amino pyridine as selective thrombin inhibitors Bioorg Med Chem Lett 1997 7 1283 1288 Pearlman D A Charifson P S Improved scoring of ligand protein interactions using OWFEG free energy grids J Med Chem 2001 44 502 511 10 The geometric mean defines the average enrichment factor as
94. e docked five sets of ligand geometries One set consists of the restraint optimized native ligands obtained from the protein preparati on procedure These ligands typically have an rmsd for non hydrogen atoms of 0 3 or less from the original PDB coordinates The second set uses MMF F 94s opti mi zed 5 versions of the native ligands Except for 1cps 1d8f 1hbv 1pro and 2cht 26 however the primary set used for assessing docking accuracy and for comparison to GOLD and FlexX consists of M MFF 94s optimized ge ometries obtained via a short M acroM odel conforma tional search starting from MMFF94s optimized ver sions of the restraint optimized native ligand geom etries In each case the optimizations used a 4r distance dependent dielectric model The fourth set consists of ligand geometries obtained using Corina while the fifth set contains geometries obtained by optimizing the Corina structures with MMF F 94s We emphasize that the initial geometry of the ligand never explicitly enters as a docked conformation How ever the conformations sampled by Glide s conformation generator depend on the input bond lengths and bond angles because these variables are not optimized F ur thermore the fact that the potential energy landscape has multiple minima and that finite sampling is done by Glide means that different solutions can be obtained from different starting points even when they are very dose to one another There is also a d
95. e finished the setup you should see a path in the title bar of the main window that ends in grids 2 1 Importing the Prepared Structures The complex for this exercise is actually in two files one containing the receptor and one containing the ligand 1 Click the Import structures toolbar button x Em The Import panel is displayed 2 From the Files of type menu ensure that Maestro is chosen 3 Click Options The Import Options dialog box opens 4 Ensure that Import all structures Replace Workspace and Fit to screen following import are all selected 5 From the Include in Workspace option menu ensure that First Imported Structure is cho sen 6 Click Close in the Import Options dialog box 7 In the Import panel navigate to the structures directory and select the file 1fjs prep recep mae gz Glide 5 5 Quick Start Guide Chapter 2 Receptor Grid Generation 10 11 12 13 14 15 Click Open The prepared protein is displayed in the Workspace The protein structure is displayed in ribbon representation The structure includes solvent molecules glycerine and ions Ca and CI but does not include the ligand Next import the ligand structure file by clicking the Import structures toolbar button mS E The Import panel is displayed From the Files of type menu ensure that Maestro is chosen Click Options The Import Options dialog box opens Deselect Replace Workspace and Fit to screen
96. e imposed based on physically reasonable values of the various parameters The present set of parameters was in fact determined by a heuristic approach a small number of XP Glide Methodology and Application paradigmatic test cases were identified for each type of term initial values were fit to yield reasonable results for these parameters and the parameter set was then tested on the entire training set Problem cases were then identified by these tests and reoptimization was carried out to improve the worst outliers A fully numerical optimization protocol would quite likely improve results for the training set but it is unclear whether a corresponding improvement in the test set would result note that test set results were not generated until the current parameter set was frozen in the Glide 4 0 release As we develop a larger test set we will investigate the use of more automated optimization algorithms with likely quantitative improvements in the predictive capabilities of the scoring function 3 Structural and Binding Affinity Prediction Results for PDB Complexes A set of 268 complexes from the PDB which we have previously used to assess the docking accuracy and scoring capabilities of Glide SP have been selected of which 198 have reliable experimentally determined binding affinities as deter mined by our extensive examination of the literature for each case This set of complexes displays a wide range of active site cavit
97. e salt bridge to be made properly Thus this parameter which normally is 2 0 kcal mol was set to 1 0 kcal mol for comparison of predicted binding affinities to experiment Other such average effects could be defined but due to the limited experimental data considered in this paper no further attempts were made to determine additional param eters Tables 6 and 7 present results for both 4 0 XP and 4 0 SP Glide for structural RMSDs taken with respect to the refined structure of the native ligand generated by our standard protein preparation procedure using heavy atoms only and for binding affinities of the various complexes discussed herein Table 6 further divides the comparison according to whether the XP complex is roughly correctly docked that is has a structural RMSD of 2 5 or less Over all 198 complexes examined the average RMSD in predicted binding affinity for the 4 0 XP scoring function is 2 3 kcal mol and the average unsigned error is 1 8 kcal mol When only well docked ligands are examined the average RMSD for the 4 0 XP scoring function is 1 7 kcal mol and the average unsigned error is 1 3 kcal mol This is a significant improvement over the perfor mance of the 4 0 SP scoring function and is also an improvement over the performances of other empirical scoring functions in the literature of which we are aware In only 1196 15 of 136 of well docked ligands were errors greater than 3 kcal mol suggesting at least when
98. e top scoring structures penalties are assessed as discussed above and the full XP scoring function is computed At this stage structures with the largest contributions from terms that promote binding may have penalties of various types The next stage of the algorithm critical to obtaining suitable results is to attempt to evade penalties by regrowing specific side chains from such poses The side chain that is the cause of the penalty can be targeted and by focusing on this region of the ligand exclusively a much larger number of candidate structures covering this region of phase space can be retained and minimized The algorithm results in a significant increase in the density of poses and locates penalty free structures when possible despite the fact that the penalty terms are not at present encoded in the energy gradient Finally a single pose is selected based on a scoring function that combines weighted Coulomb van der Waals protein ligand interaction energies the terms favoring binding affinity and the various penalty terms XP Glide Parameterization Philosophy and Implementa tion The novel terms that we have described above have been developed via a combination of reasoning from basic physical chemistry principles and examining a large set of empirical data as discussed further below Because the terms are calculated via fast empirical functions as opposed to rigorous atomistic simulations extensive parameterization is
99. echanics energy or GlideScore alone A final and very important issue is that the scoring funcion particularly the molecular mechanics compo nent needs to be modified in order to accommodate the fact that the protein structure used for docking will not in general be optimized to fit a particular ligand We have found that the most severe problem when docking a library of ligands into a single rigid receptor structure is the inability of some actives to fit into the protein cavity because the cavity is too small Therefore we typically scale down the van der Waals radii of selected e g nonpolar protein and or ligand atoms to create additional space in the binding pocket Our database enrichment studies show that this approach is effective in that context 16 Better enrichment can often be obtained by tuning the scaling parameters for a given receptor 1 but the default values are suitable for routine use 3 Scoring Function The starting point for Glide scoring is the empirically based ChemScore function of Eldridge et al 18 which can be written as AG ping Co Ciipo firi Chona Ar h o C meta ftu is C 4 H rotb 1 The summation in the second term extends over all ligand atom receptor atom pairs defined by ChemScore as lipophilic while that in the third term extends over all ligand receptor hydrogen bonding interactions In eq 1 f g and h are functions that give a full score 1 00 for distances or angles th
100. ed for assaying the top 2 5 and 10 of the ranked database a Bert site b lerr site on going from 1kim a pyrimidine site to one of the purine sites 2ki5 1ki2 1ki3 This rotation essentially exchanges the terminal NH2 and C O groups and means that purines cannot dock properly into a pyri midine site nor can pyrimidines dock properly into a purine site in each case the geometry that is correct for the parent site has an acceptor acceptor and or a donor donor clash in the alternative site This incom patibility results in larger rms deviations for cross docking as reported by Rognan and co workers 9 by J ain 2 and by us Nevertheless the misdocked purines find many favorable interactions and score well Table S1 Supporting Information because the terms in the standard precision version of GlideScore 2 5 that penal ize breaches of complementarity are too small to sig nificantly penalize the misdocked purine structures In contrast Extra Precision docking imposes larger pen alties for docking mismatches and therefore is more likely to be capable of rejecting ligands that do not dock properly into the site 2 Estrogen Receptor 3ert lerr The target proteins for the estrogen receptor ER screen are the 3ert receptor site studied by Rognan and co workers and the lerr site used by Stahl and Rarey 4 the native ligands are 4 hydroxytamoxifen and raloxifene respec tively Both sites are open enough to dock an
101. ed in the literature 6 However the methods in these papers are based on continuum solvation approaches For computing protein ligand binding affinities the role of indi vidual waters can be critical and continuum models often provide poor results in treating bound waters in a confined cavity Therefore we have chosen to implement a crude explicit water model that can be rapidly evaluated yet captures the basic physics of solvation within the confines of the protein ligand complex active site region The approach employed is to use a grid based methodology and add 2 8 A spheres approximating water molecules to high scoring docked poses emerging from the initial round of XP docking In principle this methodology is similar to that used in the GRID program though algorithmic details have been optimized to achieve speed critical in the present application The CPU time for water addition to a single pose is 3 8 CPU seconds AMD Athelon MP 1800 processor running Linux on average depending on ligand size Once waters have been added statistics are tabulated with regard to the number of waters surrounding each hydrophobic polar and charged group of the ligand and active site of the protein When a polar or charged ligand or protein group is judged to be inadequately solvated an appropriate desolvation penalty is assessed Additionally the environment of each active site water is itself probed to search for cases in which water
102. eginning and ligands with less favorable interactions appear near the end The Glide Pose Viewer panel can also be used to visualize contacts and hydrogen bonds between ligand and receptor molecules or to write structure files containing one or more ligand poses Reference Ligand A user specified structure whose ligand receptor docking score will be compared with all other docked ligands Rigid Docking A job type in which only supplied conformations of the specified ligands will be docked scored and displayed in a pose view file This job type is useful if you have already performed a conformational search on the ligands that you want to dock Glide 5 5 Quick Start Guide 120 West 45th Street 29th Floor 101 SW Main Street Suite 1300 8910 University Center Lane Suite 270 New York NY 10036 Portland OR 97204 San Diego CA 92122 Zeppelinstra e 13 Dynamostra e 13 Quatro House Frimley Road 81669 M nchen Germany 68165 Mannheim Germany Camberley GU16 7ER United Kingdom SCHR DINGER Docking with GLIDE Summary of Main Steps Project New give a name Table gt Import gt Structures select PDB file By using DUPLICATE creation of new entries containing macromolecule cofactor if any ions water molecules to be kept Icons on the left i gt A Color by Chain Name Palette icon left bottom X Delete chain B red X X Center view EN 4 Color by Element v Ribbon Display side chains
103. either is as effective as Glide 5 Sensitivity to vdW Scaling Parameters As noted previously Glide by default leaves the protein radii unchanged but scales the nonpolar ligand radii by 0 8 we refer to this as 1 0 0 8 scaling Scale factors smaller than 1 0 make the protein site roomier Enrichment Factors in Database Screening 100 80 60 100 b 10 0 an mm 5 2 60 40 20 VA 2 er a No 0 A A S V d G9 TF c er NU OE WE XT up Figure 11 Percent of actives recovered by Glide 1 8 2 0 and 2 5 and by GOLD FlexX and DOCK for assayingthe first 296 596 and 1096 of the ranked database using the datasets of Rognan and co workers gt a thymidine kinase 1kim b estrogen receptor 3ert 100 Glide 2 5 80 Glide 2 0 Glide 1 8 60 GOLD 1 1 FlexX 1 8 40 DOCK 4 01 20 ok 1 2 3 5710 2030 50 100 100 Glide 2 5 80 Glide 2 0 Glide 1 8 60 GOLD 1 1 FlexX 1 8 40 DOCK 4 01 20 E 23 5710 2030 50 100 Figure 12 Percent of known actives found y axis vs percent of the ranked database screened x axis for Glide 2 5 green Glide 2 0 blue and Glide 1 8 red and for GOLD purple FlexX orange and DOCK black using the datasets of Rognan and co workers 59 a thymidine kinase 1kim b estrogen receptor 3ert by moving back the surface of nonpolar regions of the protein and or ligand These adjust
104. en bonds between the protein and the ligand The physical argument is that the organization of water molecules to effectively solvate a structure of this type in the confined geometry of the active site can be even more problematic than that for the single hydrogen bond situation described above However this will occur only if the waters involved in such solvation are in a challenging hydrophobic environment with hydrophobic groups on two sides The coupling of the special hydrogen bond identification with the hydrophobic enclosure motif is critical if false positives are to be rejected Correlated hydrogen bonds are routinely formed in docking with highly solvent exposed backbone pairs but there is no evidence from the experimental data we have examined that such structures contribute to enhanced potency Pair correlated hydrogen bonds are defined as a donor acceptor donor donor or acceptor acceptor pair of ligand atoms referred to as ligand atom pair that are separated by no more than one rotatable bond hydroxyl groups count as nonrotatable in this calculation Several restrictions on the types of pairs that can be considered are made as detailed in Table 1 If the ligand atoms of the pair individually have zero net formal charge they must satisfy the following hydrophobicity criterion to achieve a special hydrogen bond reward First the ligand atoms must be part of the same ring or be directly connected to the same ring Assuming t
105. ependence on input ring conformation though Glide by default generates and docks alternative conformations of saturated or partly saturated five and six membered rings when it deems them to be energetically accessible We generated the conformationally optimized ligand geometries cited above to make sure that no memory of the cocrystal lized pose influenced the docking results All results are for flexible dockings carried out using Glide s internal conformation generator With the ex ception of terminal CH3 NH3t and NH2 groups all rotatable bonds are treated as optimization variables n addition as noted above alternative ring conforma tions are considered for saturated or partly saturated five and six membered rings and inversion at nitrogen is performed for asymmetric trigonal nitrogen centers in compounds such as sulfonamides The reported rms deviations in coordinate positions are based on heavy non hydrogen atoms as is also done by GOLD and FlexX 8 and are computed relative to the coordinates of the restraint opti mized native ligand structures ob tained from the protein preparation procedure Table 2 summarizes the rms errors obtained as a funcion of theflexibility of theligand As expected rms errors and CPU times increase with ligand flexibility Method and Assessment of Docking Accuracy Table 3 Comparison of rms Deviations for Flexible Docking by Glide and GOLD lt 10 rotatable bonds lt 20 rotatable bo
106. epresent global and early enrichment respectively Though 7096 re covery is arbitrary we feel this is a realistic standard for docking into a rigid protein site given that such a site is unlikely to be properly shaped to house all the known actives when the site is relatively plastic For usein virtual screening it can becrucial to concentrate as many active ligands as possible in the topmost portion of the ranked database Because the present screens use only 1000 ligands however 2 20 ranked positions or so is about the smallest percentage we can examine given that we have roughly 10 30 known actives to place We also report enrichment factors EF 596 and EF 10 for Glide 2 5 Note that the max imum attainable enrichment factors are 50 20 and 10 respectively for EF 2 EF 5 and EF 10 Also listed are average enrichment factors computed using a generalized geometric mean that weights the smaller enrichment factors more heavily 1 For example the geometric mean of 1 and 25 is 5 not 13 This weighting 1752 journal of Medicinal Chemistry 2004 Vol 47 No 7 Halgren amp al Table 1 Comparison of Enrichment Factors for Glide 1 8 Glide 2 0 and Glide 2 5 EF eq 3 EF eq 2 EF eq 2 70 recovery 2 of database 5 10 screen site GS 1 8 GS 2 0 GS 2 5 GS 1 8 GS 2 0 GS 2 5 GS 2 5 GS 2 5 thymidine kinase tk 1kim 4 2 7 6 17 2 0 0 10 0 25 0 12 0 9 0 tk pyrimidine ligands lkim 4 5 6 7 20 4 0 0 7 1 28
107. er than 10 uM As suggested above we assume that a random database will contain relatively few compounds with potency greater than this value Given the process defined above a key objective of the paper is to demonstrate the improvement that is obtained from employing the new terms defined above Comparison of version 4 0 XP Glide is made with the following 1 Version 4 0 SP Glide which has been optimized using a similar training set The 4 0 SP Glide scoring function includes some XP terms but with very small coefficients In addition to terms such as those in ChemScore from which SP Glide was originally derived this scoring function also includes contribu tions from the Coulomb and vdW protein ligand interaction energies and from Schr dinger s active site mapping technol ogy 2 A preliminary version of XP Glide v2 7 This version of XP had a nonoptimal set of penalty terms an initial version of the hydrophobic enclosure term a crude representation of the special hydrogen bond reward term without the crucial coupling to the hydrophobic enclosure term and a sampling methodology significantly inferior to that in Glide XP 4 0 This comparison enables assessment of the impact of the improved sampling and scoring methodologies used in Glide XP 4 0 Protein and Ligand Preparation Because the XP Glide scoring function is based on enforcement of physical chemical principles to a much greater degree than is employed in many other
108. es The best result obtained by a few of the 21 combinations of 3 docking engines and 7 scoring functions found 8 of 10 actives in the first 8 1096 of the ranked database Given that only 0 8 1 0 actives would be found by chance this performance Halgren amp al Table 2 Comparison of Enrichment Factors EF 70 for Glide GOLD FlexX and DOCK for the Thymidine Kinase Receptor 1kim and the Estrogen Receptor 3ert Using Data Sets of Rognan and Co workers gt 6 EF 70 70 recovery of known actives Glide Glide Glide GOLD FlexXP DOCK Screen 1 8 2 0 2 5 1 1 1 8 4 01 thymidine 42 117 193 8 2 11 1 3 0 kinase 1kim estrogen 70 0 721 471 28 5 8 9 6 7 receptor 3ert av geometric 17 1 29 0 302 15 3 9 9 4 5 mean a Equation 3 Reference 6 corresponds to an enrichment factor of roughly 10 The best single models were DOCK with PMF scoring FlexX with PMF scoring and GOLD with GOLD scoring Many models performed poorly however F or example when GOLD was used as the docking engine Chem Score the DOCK energy score and Fresno all found just one active in the first 60 of the database whereas six would be found by chance Rognan and co workers also found that some of the docking method scoring functions combinations did poorly for the estrogen receptor though others did well Our calculations used Rognan s receptor and ligand preparations and employed the 0 9 protein 0 8 ligand scaling of nonpolar vdW
109. esented in the present publication These terms are relatively small compared to the enclosure and charged charged terms discussed above but can have a nontrivial impact in specific cases For example the pi cation term for which a reward of 1 5 kcal mol is assigned is important for the acetylcholinesterase test case discussed below Journal of Medicinal Chemistry 2006 Vol 49 No 21 6185 Terms that Penalize Binding in the XP Scoring Function The most important physical effects that oppose binding are strain energy of the ligand protein or both loss of entropy of ligand and protein and desolvation of the ligand or protein The penalty terms developed are targeted in all three areas although terms addressing strain energy and entropic loss do not necessarily represent a significant advance as compared to previous terms described in the literature In developing the penalty terms some fundamental limitations arise from the rigid receptor approximation and the use of empirical scoring functions rather than full energy expressions One consequence is that in our view it is not possible to completely reject false positives with an empirical approach However significant improvements are possible as compared to previous efforts as we shall demonstrate below Water Scoring Rapid Docking of Explicit Waters Eaesolv A number of approaches to incorporating desolvation penalties into a high throughput docking code have been present
110. ethe raw Corina structures The last comparison shows that it helps to preminimize Corina derived ligands with a force field such as MMFF 94s MMFF 94s preminimization can be performed fairly easily on a large database with the premin script provided with Glide These minimizations employ a 4r distance dependent dielectric model and use M acroM od Method and Assessment of Docking Accuracy el s efficient truncated Newton minimizer Processing times are a few tenths of a second per ligand on an AMD Athelon MP 1800 processor running Linux The pro cedure keeps track of instances in which the structure recognition e g atom typing or minimization fails and automatically submits a series of MacroM odel jobs that ultimately collect the good minimized and bad un treatable structures The user can examine and fix the bad structures or can discard them 5 Discussion and Conclusions This paper has described the new computational algorithms for docking and scoring we have developed for Glide and has evaluated the performance of these algorithms in predicting binding modes over a wide range of cocrystallized structures While significant errors in binding mode prediction are found in some test cases robustness has clearly been qualitatively im proved compared to widely used alternative packages such as GOLD and FlexX Docking accuracy is also better than that reported for the recently introduced Surflex method 30 Further improvements
111. etter than its predecessors at identifying active ligands Because of improvements to the more difficult screens both global and early enrichment have tripled since the release of Glide 1 8 The CDK 2 and p38 screens are still prob lematic but thymidine kinase now does well and thrombin HIV protease thermolysin Cox 2 and HIV RT all have improved substantially Figure 1 summarizes Glide s ability to rank active ligands in the first 296 596 and 1096 of the scored and ranked database In many though not all cases a significant fraction of the actives are found in the top 296 of the database Finally Figure 2 displays the percent of known actives recovered as a function of the percent of the ranked database sampled for Glide 1 8 2 0 and 2 5 This complementary view also shows that Glide 2 5 performs exceedingly well for many of the targets and also highlights cases in which further improvement is particularly desirable Detailed Results for Database Screens The re mainder of this section describes the individual data base screens and presents graphical depictions of en richment for Glide 1 8 2 0 and 2 5 Detailed listings of the ranks of the active ligands their GlideScore values their hydrogen bonding scores and their Coulomb vdW interaction energies with the protein site are available in Supporting Information Tables SI S15 1 Thymidine Kinase 1kim Rognan and co workers studied the binding of 10 known thymidi
112. eveloped along with a correspondingly more elaborate scoring scheme However the present relatively simple scheme appears to work remarkably well at least at the level of discriminating active from inactive compounds as opposed to rank ordering which we have not yet examined in detail in a wide variety of test cases Special Charged Charged Hydrogen Bond Motifs Enb cc moti We have identified the following features that signal enhanced binding affinity from charged charged hydro gen bonds 1 The number of waters surrounding the protein component of the salt bridge Charged groups that are fully exposed to solvent are unlikely to participate in enhanced charged charged hydrogen bonding because the cost of displacing the solvent is simply too large Solvent exposure is calibrated using our water scoring code described later in this section see Eaesow by examining the number of waters in the first two shells surrounding the charged protein group 2 The number of charged charged hydrogen bonds made by the charged ligand group Three different types of salt bridge structures have been observed a Monodentate single hydro gen bond between the ligand group and a protein group b Bidentate two hydrogen bonds between the ligand group and a protein group An example of a bidentate salt bridge occurs in the lett structure of thrombin between a positively charged amine group and a recessed Asp 189 carboxylate in the releva
113. ex that is more elongated than most though not the most generous Toinvestigate the sensitivity of the docking to the choice of receptor site we chose the slightly more open laql receptor as a second site In each case we used default 1 0 0 8 scaling Figures 2d e and 5 show that Glide 2 5 outperforms its predecessors for the Idm2 site and greatly outper forms them for laql The Glide 2 5 rankings and scorings are given in Tables S4 and S5 Supporting I nformation 4 p38 MAP kinase 1a9u 1bl7 1kv2 The p38 active site is particularly prone to alter its shape upon ligand binding Therefore we studied three different PDB receptor structures 1a9u 1bl7 and 1kv2 The 1kv2 site exhibits a particularly large alteration of the ligand free structure in that a long loop undergoes a substantial change in conformation when its native ligand binds 45 This screen employs 14 known p38 binders supplied by a colleague from the biotechnology industry in addition to the cocrystallized ligands from the 1a9u 1bl7 and 1kv2 structures The plasticity of the p38 active site makes it difficult to dock a large number of active compounds properly into any single version of the receptor structure and hence leads to relatively small values of global enrich ment such as EF 70 Figures 2f h and 6 show that the 1kv2 siteis the most amenable onefor the particular selection of active compounds used in this study Glide 2 5 achieves relatively little improvement
114. exceeds a target value the user specifies the default for which is 0 3 The last structure having an rms deviation smaller than the target value is then selected We perform these mini mizations with either the Impact or Macromodel gt protein molecular modeling codes When M acroM odel is used MMFF94s5 can be employed instead of OPLS AA Our experience is that this or an equivalent prepara tion procedure is important for attaining accurate docking with Glide It is essential that physically un tenable steric clashes often found in crystallographically determined protein sites be annealed away sothat the native ligand and others can yield favorable vdW interaction energies for properly docked structures It is also important that protonation states and hydrogen bonding patterns be correct Acknowledgment This work was supported in part by grants to R A F fromthe NIH Grants P41 RR06892 and GM 52019 Supporting Information Available Table S1 listing detailed results for docking ligands from 282 cocrystallized complexes into the protein sites with Glide 2 5 This ma Method and Assessment of Docking Accuracy terial is available free of charge via the Internet at http pubs acs org References 1 Jones G Wilett P Glen R C Leach A R Taylor R Development and validation of a generic algorithm and an empirical binding free energy function J Mol Biol 1997 267 727 748 Rarey M Kramer B Leng
115. f anionic ligand functional ity to metal centers in metalloproteases n addition Glide 2 5 counts just the single best interaction when two or more metal ligations are found We set the coefficient to 2 0 kcal mol a value we believe to be reasonable though the parameter refinement would have preferred an even more strongly negative value Third we assess the net charge on the metal ion in the unligated apo protein generally straightforward via examination of the directly coordinated protein side chains If the net charge is positive the preference for an anionic ligand is incorporated into the scoring J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1741 Table 1 OPLS AA Interaction Energies with Full and Modified Charge Distributions for Ionic Centers and Groups e 2r full reduced system charges charges charged charged Zn MeOPO2z AILI 16 8 Zn MeO PO 70 8 14 5 Zn M 2PO2 68 0 13 4 Zn acetate 80 8 14 4 Mg acetate 90 0 16 2 Mn acetate 71 7 12 6 Zn CH3S7 48 7 11 5 NH4 acetate 24 8 7 1 Me guanidinium acetate 31 6 11 8 benzamidinium acetate 27 3 9 2 His acetate 22 3 7 4 Me guanidinium M eO 2PO 27 7 12 0 charged polar Zn H20 23 8 10 3 NHa4 H20 1 2 5 6 H2O acetate 8 6 4 9 Me guanidinium H 20 8 1 6 4 benzamidinium H20 6 9 48 His H20 7 2 6 5 H20 MePO2 7
116. f known drugs and other compounds identified in drug discovery projects Be cause our colleage intentionally chose the compounds to be relatively small their average molecular weight is 337 they too are smaller than typical drugs and investigational compounds which judging from Oprea s survey of the MDDR database average about 410 or 420 in molecular weight A third set of average molecular weight 350 the pc 350 set consists of 1000 represen tative compounds drawn from a 1 million compound database of purchasable compounds recently assembled by Schr dinger Finally two sets of 1000 druglike compounds of average molecular weight 360 and 400 the dl 360 and dl 400 sets were also drawn from the million compound database For the pc and dl datasets neutral database compounds were first modified by FirstDiscovery s ionizer utility to protonate or deproto nate ionizable functional groups subject to limits of 2 on the net charge and to a total of no more than four charged groups to yield ionic states likely to be present in measurable concentration between pH 5 and 9 This is to allow for shifts in pKa induced by the protein site The dl sets were selected to mimic the property distri bution of the drug lead set by using a precursor to the FirstDiscovery ligparse facility The FirstDiscovery J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1747 Table 10 Properties of 1000 Compound Ligand Databases Used
117. ficient for the great majority of ligands to enable an adequate assessment of the scoring function to be made In our optimization protocol we employed docked protein ligand complexes retaining only those with protein ligand contacts that predominantly agree with those in crystal struc tures In this fashion we avoid corrupting the fitting process with irrelevant data such as would be provided by a grossly incorrect pose yet include a realistic level of variation in the input structure This is particularly critical in optimization of the penalty terms If the sampling algorithm cannot avoid incorrect penalties in self docking it is unlikely to be able to do so in a much more challenging cross docking situation By incorporating docked poses of PDB complexes into the opti mization process the penalty function can be tuned to improve the agreement with experimental binding affinities while avoid ing inappropriately penalizing active compounds keeping in mind that there are also cases where the penalty terms are in fact appropriate Journal of Medicinal Chemistry 2006 Vol 49 No 21 6187 Before discussing binding affinity predictions a key point that has generally been neglected previously should be noted An empirical scoring function that considers only protein ligand interactions with no a priori information concerning the apo structure of the protein cannot by definition take into account the reorganization energy of the
118. following import Click Close in the Import Options panel In the Import panel navigate to the structures directory and select the file 1fjs prep lig mae gz Click Open The prepared ligand is displayed in the Workspace in tube representation 2 2 Defining the Receptor The receptor structure used for grid generation is taken from the Workspace so you need to exclude the ligand atoms from consideration as part of the receptor 1 From the Applications menu in the main window choose Glide gt Receptor Grid Genera tion The Receptor Grid Generation panel opens with the Receptor tab displayed In the Define receptor section ensure that Pick to identify ligand and Show markers are selected and that in the option menu Molecule is chosen In the Workspace pick an atom in the ligand molecule Dark green markers appear on the ligand In the Van der Waals radii scaling section ensure that Scaling factor is set to the default value of 1 00 no scaling Glide 5 5 Quick Start Guide Chapter 2 Receptor Grid Generation Receptor Grid Generation EEE Receptor Site Constraints Rotatable Groups m Define receptor If the structure in the Workspace is a receptor plus a ligand you must identify the ligand molecule so it can be excluded from the grid generation Molecule 3 Show markers Van der Waals radius scaling To soften the potential for nonpolar parts of the receptor you can scale the van
119. gands in the Training Set screen Ephobic pair Erp pair Enb_nn_motit Enb cc motif Ehya enclosure Ep Ebina acetylcholinesterase 121 0 3 0 0 19 4 4 zd 20 8 neuraminidase 3 8 0 7 0 4 3 6 0 0 0 0 8 6 factor Xa 8 3 0 8 L8 1 2 0 8 0 8 14 1 human p38 map kinase 102 1 6 0 3 0 0 2 6 0 0 14 7 p38 map kinase 14 12 0 0 0 0 0 9 0 0 9 5 HIV RT 91 1 0 14 0 0 4 2 0 0 137 cyclooxygenase 2 8 9 0 3 0 0 0 0 3 2 0 0 125 human cyclin dep kinase 1 8 2 1 32 0 0 1 5 0 0 14 7 thrombin 8 4 17 0 0 30 0 4 0 0 13 1 HIV 1 protease 10 2 2 6 0 2 0 0 0 0 0 0 13 1 human estrogen receptor 10 6 12 0 5 0 0 2 0 0 0 14 3 Ick kinase 8 1 1 4 L3 0 0 0 0 0 0 10 8 EGRE tyrosine kinase 6 7 13 15 0 0 0 0 0 0 9 5 thermolysin 8 7 2 3 0 0 0 3 0 1 0 0 11 4 thymidine kinase II 2 3 3 0 0 0 14 0 0 12 6 Ephobic_pair iS the pair lipophilic term eq 1 Enb_pair is the Chemscore like pair hydrogen bond term Enb_nn_motif is the term for neutral neutral hydrogen bonds in a hydrophobically enclosed environment Epb_cc_motif is the term for special charged charged hydrogen bonds Enya_enclosure is the hydrophobic enclosure reward Ep is the pi stacking pi cation reward and Ebina is the sum of all terms in eq 3 that account for favorable binding affinities different types of active sites and binding motifs A rough classification of
120. gands or to other docking codes However we have been given access to the thymidine kinase and estrogen receptor datasets employed by Bissantz Folkers and Rognan To whom correspondence should be addressed Phone 646 366 9555 extension 106 Fax 646 366 9550 E mail halgren schrodinger com t Schr dinger L L C NY Columbia University Schr dinger L L C OR and offer comparisons tothe results they published for GOLD FlexX and DOCK The paper is organized as follows In the section 2 we characterize our data sets and protocols for evaluat ing database enrichment This section describes the receptors and ligands to be used discusses certain issues concerning preparation of the receptor most importantly the use of reduced van der Waals radii which is essential to achieve reasonable results in some cases and defines the quantitative measures used to assess performance in database screening Section 3 presents enrichment factors obtained using default parameters and describes the individual screens Com parisons to published results for GOLD FlexX and DOCK for the thymidine kinase and estrogen receptors are then presented in the fourth section and Glide s sensitivity to the choice of certain input factors is explored in section 5 Finally the sixth section sum marizes the results and discusses future directions 2 Virtual Screening Protocol Ligand Databases and Receptors Used We have chosen the
121. he pair satisfies these restrictions the hydrophobicity of the hydrogen bond region is detected in a manner similar though not identical to that for a single special hydrogen bond A sum of the hydrophobic enclosure packing score described previously is made for the pair of hydrogen bonding ligand heavy atoms and the ring atoms directly connected to the ligand pair atoms If a ligand atom of the pair is not a ring atom but is connected to a ring the sum includes atoms of the ring that are nearest neighbors to the nonring ligand atom If the absolute hydrophobic sum is above a cutoff the hydrogen bonded pair is given a special reward of 3 kcal mol Note that double counting of pair and single special hydrogen bonds is avoided by checking pairs first and excluding any rewarded hydrogen bonds found from single hydrogen bond consideration Finally the special hydrogen bond rewards are linearly reduced with the quality of the hydrogen bond The hydrogen XP Glide Methodology and Application Figure 4 Staurosporine bound to human cyclin dependent kinase CDK2 The pair of correlated hydrogen bonds receives a 3 kcal mol reward while the central component of the ring is given a 3 kcal mol hydrophobic enclosure packing reward Figure 5 AG12073 bound to human cyclin dependent kinase CDK2 The three correlated hydrogen bonds receive a 4 2 kcal mol reward bond quality is measured in the sense of the pair hydrogen bond score us
122. he rules for charged charged interactions laid out in section 2 In GIuR2 an unusual set of charged ligands in which charge is distributed over a ring system are buried in the active site This system was used to calibrate the use of electrostatic terms to turn off buried charge penalties and to assign an additional contribution to binding affinity in exceptional cases as discussed in section 2 7 The acetylcholinesterase 1e66 receptor was included to incorporate the well known pi cation motif of the active compounds in the parameterization process Table 8 lists the average contribution of the hydrophobic enclosure and special hydrogen bonding terms to the scores for known active compounds that bind to each of the above targets as well as the average total score Note that the total score cannot be directly translated into the predicted binding affinity in all cases because there are a number of targets where allosteric rearrangement of the binding pocket leads to substantial protein XP Glide Methodology and Application Figure 10 SCX 001 bound to cyclooxygenase 2 The three phenyl rings obtain a 3 7 kcal mol hydrophobic enclosure packing reward for occupying the large hydrophobic pocket Figure 11 The 1lpk ligand bound to factor Xa A special salt bridge pair with ASP 189 analogous to that in thrombin Figure 8 is observed A hydrophobic enclosure packing reward of 0 8 kcal mol is achieved by the phenyl ring occupying
123. hen employed to minimize the pc and dl ligands with M MFF 94s 2 using a 4r distance dependent dielectric All compounds considered had 100 or fewer atoms and 20 or fewer rotatable bonds The properties of these ligand databases are shown in Table 10 We believe that the dl 400 set is represen tative of ligands one would expec to find in the compound collection of a pharmaceutical or biotechnol ogy company We weighted the dl 400 set the most heavily in the parametrization of GlideScore 2 5 but to broaden the parametrization we used the others as well and also included screens run with nonstandard values for the protein and ligand scaling factors The large number of screens employed 94 in all embracing 16 receptor sites should guard against overfitting and help to make database screening with Glide as tolerant as possible to variations in the ligand sets and the vdW scale factors Thelipophilic contac term in Glide 2 5 SP scoring eq 2 contributes 4 85 kcal mol to an average score of 9 33 kcal mol for active compounds induded in the database screens and a second contact term the vdW interaction energy contributes 2 39 kcal mol on aver age Thus these two terms account for nearly 8096 of the total score The hydrogen bonding terms are next largest in importance their average contributions for the actives being 1 12 kcal mol representing roughly two hydrogen bonds One major change in the scoring concerns the way in
124. his Document Conventions In descriptions of command syntax the following UNIX conventions are used braces enclose a choice of required items square brackets enclose optional items and the bar symbol separates items in a list from which one item must be chosen Lines of command syntax that wrap should be interpreted as a single command File name path and environment variable syntax is generally given with the UNIX conven tions To obtain the Windows conventions replace the forward slash with the backslash in path or directory names and replace the at the beginning of an environment variable with a at each end For example SCHRODINGER maestro becomes 3SCHRODINGER maestro In this document to type text means to type the required text in the specified location and to enter text means to type the required text then press the ENTER key References to literature sources are given in square brackets like this 10 Glide 5 5 Quick Start Guide v vi Glide 5 5 Quick Start Guide Chapter 1 Getting Started This manual contains tutorials designed to help you quickly become familiar with the function ality of Glide using the Maestro interface This chapter contains a brief overview of the soft ware and some setup instructions for the tutorials The tutorial begins in Chapter 2 with the generation of grids from a prepared protein to represent the receptor for docking In Chapter 3 a set of ligands is d
125. his material is available free of charge via the Internet at http pubs acs org References 1 Friesner R A Banks J L Murphy R B Halgren T A Klicic J J Mainz D T Repasky M P Knoll E H Shelley M Perry J K Shaw D E Francis P Shenkin P S Glide A New Approach for Rapid Accurate Docking and Scoring 1 Method and Assessment of Docking Accuracy J Med Chem 2004 47 1739 1749 Halgren T A Murphy R B Friesner R A Beard H S Frye L L Pollard W T Banks J L Glide A New Approach for Rapid Accurate Docking and Scoring 2 Enrichment Factors in Database Screening J Med Chem 2004 47 1750 1759 3 Perola E Walters W P Charifson P S A Detailed Comparison of Current Docking and Scoring Methods on Systems of Pharma ceutical Relevance Proteins 2004 56 235 249 4 Krovat E M Steindl T Langer T Recent Advances in Docking and Scoring Curr Comput Aided Drug Des 2005 1 93 102 5 Kontoyianni M McClellan L M Sokol G S Evaluation of Docking Performance Comparative Data on Docking Algorithms J Med Chem 2004 47 558 565 6 Sherman W Day T Jacobson M P Friesner R A Farid R Novel Procedure for Modeling Ligand Receptor Induced Fit Effects J Med Chem 2006 49 534 553 7 Farid R Day T Friesner R A Pearlstein R A New Insights about HERG Blockade Obtained from Protein Modeling Pote
126. hough approximations and truncations are required to achieve acceptable computational speed Starting from the poses selected by the initial screen ing the ligand is minimized in the field of the receptor using a standard molecular mechanics energy function in this case that of the OPLS AA force field 7 in Friesner amp al conjunction with a distance dependent di electric model Finally the three to six lowest energy poses obtained in this fashion are subjected to a Monte Carlo pro cedure that examines nearby torsional minima This procedure is needed in some cases to properly orient peripheral groups and occasionally alters internal tor sion angles We and others have found that a conventional mo lecular mechanics energy function is a reasonable model for predicting binding modes even in the absence of solvent However it is not inadequate for ranking disparate ligands for example ligands with different net charge Therefore we have implemented a modified and expanded version of the ChemScore scoring func tion GlideScore for use in predicting binding affinity and rank ordering ligands in database screens How ever we use a combination of GlideScore the ligand receptor molecular mechanics interaction energy and the ligand strain energy to select the correctly docked pose We find that this composite scoring function which we call Emodel is much better at selecting the correct pose than is either the molecular m
127. ic and Macroscopic Hydrophobic Effects Science 1991 252 106 109 30 Regan J Breitfelder S Cirillo P Gilmore T Graham A G Hickey E Klaus B Madwed J Moriak M Moss N Pargellis C Pay S Proto A Swinamer A Tong L Torcellini C Pyrazole Urea Based Inhibitors of p38 MAP Kinase From Lead Compound to Clinical Candidate J Med Chem 2002 45 2994 3008 31 Hendsch Z S Tidor B Do Salt Bridges Stabilize Proteins A Continuum Electrostatic Analysis Protein Sci 1994 3 211 226 32 Waldburger C D Schildbach J F Sauer R T Are Buried Salt Bridges Important for Protein Stability and Conformational Specific ity Nat Struct Biol 1995 2 122 128 33 Marqusee S Sauer R T Contributions of a Hydrogen Bond Salt Bridge Network to the Stability of Secondary and Tertiary Structure in Lambda Repressor Protein Sci 1994 3 2217 2225 34 Luo R David L Hung H Devaney J Gilson M K Strength of Solvent Exposed Salt Bridges Strength of Solvent Exposed Salt Bridges J Phys Chem B 1999 103 721 136 35 Krammer A Kirchhoff P D Jiang X Venkatachalam C M Waldman M LigScore A Novel Scoring Function for Predicting Binding Affinities J Mol Graphics Modell 2005 23 395 407 36 Wei B Q Q Baase W A Weaver L H Matthews B W Shoichet B K A Model Binding Site for Testing Scoring Functions in Molecular Docking J Mol Biol
128. ids will be cal culated Generally you should make the enclosing box as small as is consistent with the shape and character of the protein s active site and with the ligands you expect to dock 2 In the Site tab ensure that the Center option selected is Centroid of Workspace ligand 3 Ensure that the Size option selected is the default Dock ligands similar in size to the Workspace ligand If you have a representative ligand in the active site the default generates an enclosing box that is large enough for most systems However if you think that conformations of active ligands may exist that are significantly larger than the cocrystallized ligand you should consider enlarging the enclosing box using the Dock ligands with length lt option Receptor Grid Generation BEE Receptor Site Constraints Rotatable Groups Enclosing box The docked ligand is confined to the enclosing box X Center Centroid of Workspace ligand selected in the Receptor tab gt Centroid of selected residues Specify Residue Supplied X Y Z coordinates Y z Size Dock ligands similar in size to the Workspace ligand Dock ligands with length lt J JA Advanced Settings Start Write Reset Close Help Figure 2 2 The Site tab of the Receptor Grid Generation panel Glide 5 5 Quick Start Guide Chapter 2 Receptor Grid Generation
129. ies and protein ligand interactions The parameters of the scoring function were simultaneously optimized to reproduce the experimental binding affinity data and yield quality enrichment factors binding affinities for the database screening tests that are discussed below An evaluation of the docking accuracy of Glide SP was presented in ref 1 and the XP results from docking MMFFs optimized ligand structures shown in Table 5 are very similar This suggests that the sources of error in docking accuracy are due to issues other than those addressed by the XP scoring function modifications In some cases near symmetry in the ligand leads to docked poses that are functionally equivalent to those in the crystal structure for example in terms of protein ligand contacts but that exhibit a large RMSD such cases are identified in Table 5 A principal source of errors in pose RMSD appears to be the charge distribution of the ligand which in a standard force field representation may not accurately distribute formal ionic charges and does not incorporate polarization effects Cho and co workers have demonstrated that low RMSD ligand poses can be reliably generated by utilizing more accurate polarized charges where the ligand charge distribution is computed in the protein environment via QM MM methods We may incorporate this methodology into future XP Glide releases for optional use The quality of structural prediction shown by XP Glide is suf
130. imal increases Our objective has been to explain these results on the basis of physical chemical principles and to develop empirical scoring terms that captured the essential physics while rejecting false positives even with imperfect docking and the neglect of induced fit effects Calculation of the hydrophobic enclosure score Ehyd enclosure is summarized below with a more detailed description of the algorithm provided in Supporting Information 1 Lipophilic protein atoms near the surface of the active site and lipophilic ligand atoms are divided into connected groups There are a set of rules specifying which atoms count as lipophilic and what delimits a group 2 For each atom in a group on the ligand lipophilic protein atoms are enumerated at various distances 3 For each lipophilic ligand atom the closest lipophilic protein atom is selected and a vector is drawn between it and the ligand atom This is the protein anchor atom for that ligand atom Vectors for all other suitably close lipophilic protein atoms are drawn to the ligand atom and their angles with the anchor atom vector are determined To be considered on the opposite side of the anchor atom the angle between vectors must exceed a cutoff value that depends on the pair distance with shorter distances requiring that the angle be closer to 180 If the angle is close to zero degrees the atom is on the same side and is at right angles to the anchor if the
131. in Database Screens cmc drug property av pdb lead pc350 di 360 dl 400 molecular weight 290 337 350 360 400 atoms 38 30 42 83 42 79 45 74 50 75 non hydrogen atoms 22 44 26 37 25 63 2813 31 28 higher row atoms 0 52 0 47 0 82 0 50 0 56 96 hydrophobic 58 52 59 12 63 67 58 94 59 12 carbons rings 153 2 09 2 33 2 23 2 51 heteroaramaticrings 0 24 0 84 0 40 0 90 0 99 rotatable bonds 5 51 5 80 5 95 6 19 6 92 amide hydrogens 0 39 0 27 0 66 0 29 0 33 neutral donors 161 1 46 1 02 1 56 1 70 charged donors 0 88 0 96 0 30 1 02 1 14 neutral acceptors 187 1 49 2 09 1 59 1 78 charged acceptors 0 68 0 42 0 11 0 45 0 49 divalent oxygens 0 55 0 68 0 78 0 73 0 82 neutral amines 0 05 0 01 0 02 0 11 0 13 acidic hydrogens 0 02 0 01 0 00 0 04 0 04 Each ligand database contains 1000 compounds b Taken from a subset of the Comprehensive Medicinal Chemistry database or from cocrystallized PDB complexes Compounds of relatively low molecular weight from the Derwent World Drug Index provided by a pharmaceutical collaborator 4 Extracted from a 1 million compound database of purchasable compounds assembled by Schr dinger in such a way as to preserve the distribution of listed properties Extracted from the million compound database in such a way as to preserve proportionately scaled values of properties of the drug lead set however the distribution of the last two listed properties was not controlled for premin utility was t
132. in the rigid receptor approximation the only solutions are a to adjust the parameters so that they allow the test case in question to escape penalization or b to accept that the active compound does not fit into the particular version of the receptor being used and to dock into multiple structures and or employ induced fit methodologies For many cases however better sampling in the relevant phase space can locate ligand geometries that are able to avoid the penalties The XP Glide sampling algorithm was explicitly designed with this objective in mind XP Glide Sampling Methodology XP Glide sampling begins with SP Glide docking as described in refs 1 and 2 but using a wider docking funnel so that a greater diversity of docked structures is obtained For XP docking to succeed SP docking must produce at least one structure in which a key fragment of the molecule is properly docked This has been the case in the great majority of systems that have been investigated The second step in XP sampling is to assign various fragments of the molecule as anchors and to attempt to build a better scoring pose for the ligand starting from each anchor Typical anchors are rings but can be other rigid fragments as well Various positions of the anchors are clustered representative members of each cluster are chosen and the growing of side chains from relevant positions on the anchor is initiated The growing algorithm proceeds one
133. ing the donor acceptor distance and the angle made by the donor heavy atom H vector and the H acceptor vector For ligand acceptor atoms in rings the extent to which the acceptor lone pair vector is aligned with the donor heavy atom H vector is evaluated A detailed description of the algorithm for scaling the special hydrogen bond reward is given in Supporting Information A substantial number of protein ligand complexes in which motifs containing correlated hydrogen bonds that satisfy the above criteria including the requisite hydrophobic enclosure have been identified A number of examples are shown below Figure 4 depicts the laql structure of staurosporine bound to human cyclin dependent kinase CDK2 This type of correlated pair is also found in a number of other kinases Some of the CDK2 actives such as AGI2073 make three correlated hydrogen bonds this structure is shown in Figure 5 A second example is streptavidin bound to the Istp structure of biotin Figure 6 with three correlated hydrogen bonds in a hydro phobically enclosed region To our knowledge no empirical scoring function has explained the exceptionally large binding affinity of streptavidin to biotin However once the correlated hydrophobically enclosed hydrogen bonding motif is recognized and assigned an appropriate score a reward of 3 kcal mol consistent with other examples the deviation between calcu Journal of Medicinal Chemistry 2006 Vol 49 No 21 618
134. ion while CDK2 actives form either a pair or a triplet and neutral actives binding to aldose reductase form a correlated triplet similar to that in the streptavidin biotin pair mentioned above TK is a small relatively polar site with the binding driven primarily by the special hydrogen bonding whereas CDK2 and LCK are medium sized and more hydro phobic binding ligands in the 400 500 molecular weight range although smaller ligands can bind as well 4 Large buried predominantly hydrophobic sites This category includes the human estrogen receptor and the 1kv2 conformation of p38 MAP kinase The estrogen receptor binds large flat steroid type molecules and exhibits a medium sized hydrophobic enclosure term with the binding driven by this term and by the pair hydrophobic score The 1kv2 active site is created by an allosteric rearrangement of the p38 activation loop It has a large hydrophobic enclosure term and many active compounds make one to two hydrogen bonds The most active compound ligand 1kv2 also makes a special hydrogen bond in the hinge region but binding is primarily driven by hydrophobic terms The p38 pocket in particular is quite large and in the absence of the hydrophobic enclosure term will display a significant number of false positives that bind alternative motifs in the cavity in database screening 5 Large open sites with relatively shallow cavities in the active site pocket This category includes thrombi
135. ions of the ligand are made of 0 3 in magnitude At each perturbed geometry the sum is recalculated and the reward is Friesner et al Table 1 conditions for not applying the special pair hydrogen bond scores ignore poor quality hydrogen bonds 0 05 Eno pair ignore pairs involving the same neutral protein atom ignore pairs involved in a salt bridge if the electrostic potential at either ligand atom is above the cutoff ignore salt bridge pairs if either protein atom is involved in a protein protein salt bridge ignore ligand donor donor pairs that come from NH where x 2 groups the nitrogen atom is not in a ring and has no formal charge ignore formally charged protein atoms with more than eight second shell waters in the unligated state ignore charged neutral hydrogen bonds unless the protein atom is in a salt bridge ignore pairs of different neutral acceptor atoms on the ligand neutral hydrogen bond pairs must satisfy ring atom and hydrophobicity environment criteria as outlined in section 2 ignore ligand hydroxyl to protein hydrogen bonds if the protein atom has zero formal charge applied if at some geometry the sum exceeds the cutoff This procedure helps to avoid discontinuities inherent in the use of a cutoff The situation described above identifies a structure in which a single hydrogen bond should be assigned a special reward A second situation occurs when there are multiple correlated hydrog
136. ired to fully occupy the hydrophobic pocket depicted in Figure 2 As indicated above the surrounding of ligand lipophilic atoms or groups by lipophilic protein atoms is referred to as hydro phobic enclosure Our contention here and in much of the following discussion of hydrogen bonding is that proper treatment of hydrophobic enclosure is the key to discrimination of highly and weakly potent binding motifs and compounds The underlying mathematical framework for describing enclo sure discussed above could be cast in other forms but the essential idea would remain unchanged Detailed optimization of the numerical criteria for recognizing enclosure and assigning XP Glide Methodology and Application Figure 2 Boehringer active for 1kv2 bound to human p38 map kinase The naphthyl group receives a 4 5 kcal mol hydrophobic enclosure packing reward a specific contribution to the binding affinity for each motif is vital to developing methods with predictive capability Improved Model of Protein Ligand Hydrogen Bonding In developing a refined model of hydrogen bonding we divide hydrogen bonds into three types neutral neutral neutral charged and charged charged The analysis of each type of hydrogen bonding is different due to issues associated with the long range solvation energy Born energy of charged groups An initial step is to assign different default values assuming optimal geometric features to each of the three
137. irst The archive is named Jobid archive tar gz and should be sent to help schrodinger com instead If Maestro fails an error report that contains the relevant information is written to the current working directory The report is named maestro error txt and should be sent to help schrodinger com A message giving the location of this file is written to the terminal window More information on the postmortem command can be found in Appendix A of the Job Control Guide On Windows machine and system information is stored on your desktop in the file schrodinger machid txt If you have installed software versions for more than one release there will be multiple copies of this file named schrodinger machid N txt where N is a number In this case you should check that you send the correct version of the file which will usually be the latest version If Maestro fails to start send email to help schrodinger com describing the circumstances and attach the file maestro error txt If Maestro fails after startup attach this file and the file maestro EXE dmp These files can be found in the following directory SUSERPROFILI GI Local Settings Application Data Schrodinger appcrash Glide 5 5 Quick Start Guide 37 38 Glide 5 5 Quick Start Guide Glossary Base Name The name entered in the Base name for grid files text box that is used to write grid files during a grid file calculation or to find pre exi
138. l 49 No 21 Table 10 Standard Enrichment Factors for Recovering 40 of the Correctly Docked Active Ligands in the Training Set enrichment factors screen v4 0 XP v2 7 XP v4 0 SP acetylcholinesterase 37 1 2 neuramididase 64 2 112 factor Xa 126 42 63 human p38 map kinase 81 58 51 p38 map kinase 35 18 11 HIV RT 36 14 18 cyclooxygenase 2 35 56 78 human cyclin dep kinase 168 168 8 thrombin 68 58 12 HIV 1 protease 90 32 112 human estrogen receptor 126 126 126 Ick kinase 12 6 EGRE tyrosine kinase 6 1 2 thermolysin 50 134 201 thymidine kinase 251 251 17 2 Active ligands have at least 10 uM activity except those for neuraminidase as described in section 4 previously The results in Table 10 include the same sets of active ligands as in Table 9 that is those whose activities are better than 10 uM and have been judged to fit more or less correctly into the specified conformation of the receptor Table 11 presents results using all active compounds with binding affinities better than 10 uM whether the binding mode is judged to be correct Results are presented for recovering 40 70 and 100 of the considered active ligands in each case This type of analysis corresponds to the approach taken by us in ref 2 as well as to other work in the literature While we believe that the analysis in Table 9 is the appropriate one to use in assessing the quality of a scoring function to be used in rigid receptor docking the result
139. la 8 7 8 4 8 4 5 1 0 39 6 01 ladf 6 2 79 7 9 11 8 3 03 9 92 lelb 9 8 6 3 6 3 7 0 5 42 4 29 laha 79 19 8 2 0 36 0 11 lelc 9 4 7 3 1 3 6 0 6 53 7 92 lake 14 0 14 0 3 9 15 45 14 95 leld 9 1 6 5 6 5 4 8 0 32 3 94 lapb 7 9 SVT E 94 0 06 0 10 lele 9 3 8 0 8 0 6 4 0 36 0 38 lapt 12 8 119 11 9 11 0 1 48 1 24 lepb 13 6 8 0 2 24 1 89 lapu 10 2 8 7 8 7 1 6 0 61 1 24 leta 3 9 3 9 35 8 69 1 85 lapv 12 3 11 2 11 2 8 5 0 63 0 48 letr 10 1 11 7 9 7 9 4 0 71 0 68 lapw 10 9 11 0 11 0 9 0 0 97 0 32 lets 11 2 13 7 11 7 11 7 1 32 1 44 latl 8 6 10 6 10 6 7 8 0 87 3 47 lett 8 0 12 6 10 6 9 5 0 62 0 58 lavd 16 4 16 4 10 4 0 82 0 55 lezq 12 3 11 4 9 4 12 6 0 75 0 21 lazm 1 e 6 2 1 89 2 51 1f0r 10 4 13 3 11 3 10 1 2 11 0 59 1b6j 10 8 18 3 18 3 15 8 2 98 0 43 1f0s 10 6 113 11 3 9 4 2 08 0 35 1b6k 11 9 14 8 12 8 13 8 1 04 1 06 1f0t 8 2 7 7 Til 9 8 0 38 0 24 1b6l 11 3 11 1 11 1 8 7 0 92 1 18 1f0u 9 8 10 8 8 8 10 5 1 56 1 54 1b6m 11 5 13 9 11 9 11 4 0 73 3 17 lfen 12 8 12 1 84 1 10 0 40 lbaf 9 5 9 5 8 3 1 17 1 08 1fh8 9 4 10 7 10 7 10 3 0 20 0 20 lbap 9 3 8 2 8 2 9 1 0 39 0 38 1fh9 8 8 6 0 6 0 53 2 11 1 99 1bbp 14 3 12 4 11 8 5 28 5 11 1fhd 9 3 8 1 8 1 8 7 5 37 0 46 Ibkm 11 6 11 6 14 6 4 77 2 36 Ifjs 1
140. lation Flexible Docking A job type in which alternate conformations for each ligand are generated during the docking process and then the interactions between the receptor and the conformers are analyzed After docking jobs are complete the conformers or poses are ranked according to their overall interaction with the receptor The results can be posted to a pose view file which can be examined using the Glide Pose Viewer panel GlideScore Glide s scoring function based on ChemScore GlideScore is used in ranking ligand poses found in docking In Liaison GlideScore is used in an alternative binding energy model Grid Files Files written by Glide during grid setup These files contain data about the prop erties of the associated receptor and are used during docking Glide 5 5 Quick Start Guide 39 Glossary 40 Ligand Centroid Used to define the enclosing box center a ligand centroid is the point whose x y and z coordinates are the mean of the minimum and maximum x y and z coordi nates of all the atoms in the ligand Pose Viewer Panel An analysis tool that displays the results of Glide docking jobs These results which are recorded in a pose view file include the ligand name pose number overall score number of contacts and other data The poses within the file are arranged in the list according to score ligands with the most energetically favorable interactions with the receptor appear at the b
141. lation_path bash ksh export SCHRODINGER installation_path This environment variable is also required to run Glide jobs 2 Change to the desired working directory cd directory name 3 Enter the following command SSCHRODINGER maestro amp The Maestro main window is displayed and the working directory is Maestro s current working directory If you are using an existing Maestro session you can change the direc tory by choosing Change Directory from the Maestro menu navigating to the appropriate directory and clicking OK Windows 1 Double click the Maestro icon on the desktop You can also use the Start menu Maestro is in the Schr dinger submenu 2 From the Maestro menu choose Change Directory 3 Navigate to the working directory and click OK Glide 5 5 Quick Start Guide Chapter 2 Receptor Grid Generation This chapter contains exercises that demonstrate how to use the Receptor Grid Generation panel to set up and start a grid file calculation job Grid files represent physical properties of a volume of the receptor specifically the active site that are searched when attempting to dock a ligand You will use the grid files calculated in this chapter to dock ligands in later Glide exer cises If you have not started Maestro start it now Before proceeding with the exercises change the working directory to the grids directory See Section 1 3 on page 3 for instructions on how to do these tasks When you hav
142. lecular docking Protein Eng 1993 6 723 732 24 Muegge I Martin Y C A A general and fast scoring function for protein ligand interactions a simplified potential approach J Med Chem 1999 42 791 804 25 Rognan D Laumoeller S L Holm A Buus S Tschinke V Predicting binding affinities of protein ligands from three dimensional coordinates Application to peptide binding to class major histocompatibility proteins J Med Chem 1999 42 6450 4658 26 Wang R Liu L Tang Y SCORE a new empirical method for estimating the binding affinity of a protein ligand complex J Mol Modd 1998 4 379 384 J M030644S J Med Chem 2006 49 6177 6196 6177 Extra Precision Glide Docking and Scoring Incorporating a Model of Hydrophobic Enclosure for Protein Ligand Complexes Richard A Friesner Robert B Murphy Matthew P Repasky Leah L Frye Jeremy R Greenwood Thomas A Halgren Paul C Sanschagrin and Daniel T Mainz Department of Chemistry Columbia University New York New York 10027 Schr dinger Limited Liability Company 120 West 45th Street New York New York 10036 Schr dinger Limited Liability Company 101 SW Main Street Portland Oregon 97204 Received December 16 2005 A novel scoring function to estimate protein ligand binding affinities has been developed and implemented as the Glide 4 0 XP scoring function and docking protocol In addition to unique water de
143. lecular docking to induced fit effects Applica tion to thrombin thermolysin and neuraminidase J Comput Aided Mol Des 1999 13 547 562 9 McGann M Almond H Nicholls A Grant J A Brown F Gaussian docking functions Biopolymers 2003 68 76 90 10 Venkatachalam C M J iang X Oldfield T Waldman M J LigandFit a novel method for the shape directed rapid docking of ligands to protein active sites J Mol Graphics M oddl 2003 21 289 307 11 Charifson P S Corkery J J Murcko M A Walters W P Consensus scoring A method of obtaining improved hit rates from docking databases of three dimensional structures into proteins J Med Chem 1999 42 5100 5109 12 Bissantz C Folkers G Rognan D Protein based virtual screening of chemical databases 1 Evaluation of different docking scoring combinations J Med Chem 2000 43 4759 4767 13 Stahl M Rarey M Detailed analysis of scoring functions for virtual screening J Med Chem 2001 44 1035 1042 14 Schr dinger L L C New York The Glide 2 5 calculations used FirstDiscovery version 2 5021 which was released in J une 2003 15 The results for the GOLD test set are available at http www ccdc cam ac uk prods gold rms_tab html 16 Halgren T A Murphy R B Friesner R A Beard H S Frye L L Pollard W T Banks J L Glide A new approach for rapid accurate docking and scoring 2
144. ligand 16088 The hydrogen bonds are displayed as yellow dotted lines and annotated with their lengths 8 Click the Electro cell for ligand 16088 The amidine nitrogens are displayed in ball and stick to indicate their contribution to electrostatic rewards Glide 5 5 Quick Start Guide 33 Chapter 4 Examining Glide Data 34 9 Click the x Stack cell for ligand 16088 The aromatic groups in the protein that contribute to the x stacking reward are displayed in gray in CPK and the ligand aromatic group is displayed in ball and stick 10 Click the RotPenal cell for ligand 16088 The bonds that contribute to the rotatable bond penalty are displayed as tubes 11 Close the Glide XP Visualizer panel 12 Clear the Workspace deleting the scratch entry 4 6 Finishing the Exercises Close the scratch project you are working in Because you have written the output structure files to your directory tree you do not need to save the scratch project or Workspace structures Click OK to delete any scratch entries From the Maestro menu choose Quit and click Quit do not save log file For more information about Quit panel options and maestrolog cmd files click Help instead Glide 5 5 Quick Start Guide Getting Help Schr dinger software is distributed with documentation in PDF format If the documentation is not installed in SCHRODINGER docs on a computer that you have access to you should install it or ask your syste
145. m administrator to install it For help installing and setting up licenses for Schr dinger software and installing documenta tion see the Installation Guide For information on running jobs see the Job Control Guide Maestro has automatic context sensitive help Auto Help and Balloon Help or tooltips and an online help system To get help follow the steps below Check the Auto Help text box which is located at the foot of the main window If help is available for the task you are performing it is automatically displayed there Auto Help contains a single line of information For more detailed information use the online help If you want information about a GUI element such as a button or option there may be Balloon Help for the item Pause the cursor over the element If the Balloon Help does not appear check that Show Balloon Help is selected in the Maestro menu of the main window If there is Balloon Help for the element it appears within a few seconds e For information about a panel or the tab that is displayed in a panel click the Help button in the panel or press F1 The help topic is displayed in your browser For other information in the online help open the default help topic by choosing Online Help from the Help menu on the main menu bar or by pressing CTRL H This topic is dis played in your browser You can navigate to topics in the navigation bar The Help menu also provides access to the manuals including
146. maining controls in the Define core section become available 14 Under Core atoms select SMARTS pattern 15 In the main window from the Undisplay toolbar button menu choose Nonpolar hydro gens f 16 Rotate the structure so that you can clearly see the three six membered rings Glide 5 5 Quick Start Guide 25 Chapter 3 Ligand Docking 26 17 18 19 Ensure that the Workspace selection button is selected indented and displays an A for picking atoms x Select the three six membered rings with their ether linkages and the amidine group on the terminal ring Do not include the carboxyl on the middle ring or the hydroxyl on the terminal ring You can drag to make the first selection then hold down the SHIFT key and drag or click to add atoms to the selection The selected atoms are marked in yellow rather than in purple In the Ligand Docking panel click Get From Selection The Smarts pattern text box is filled in with the pattern that corresponds to the atoms selected in the Workspace The markers on the Workspace selection turn green Finally the ligands to be docked need to be selected A different set is used from the set used for the previous runs 20 21 22 23 24 25 In the Ligands tab ensure that File is selected Click Browse A file selector is displayed Ensure that Files of type is set to Maestro Navigate to the tutorial structures directory choose sar series m
147. membered rings and number of asymmetric trigonal nitrogen centers in compounds such as sulfonamides core is what remains when each terminus of the ligand is severed at the last rotatable bond as is indicated in the figure the directly attached atom of each rotomer group is also considered to be part of the core Carbon and nitrogen end groups terminated with hydrogen CH3 NH2 NH3 are not considered rotatable because their conformational variation is of little inter est Each core region is represented by a set of core conformations the number of which depends on the numbers of rotatable bonds conformationally labile five and six membered rings and asymmetric pyramidal trigonal nitrogen centers in the core As Table 8 shows this set typically contains fewer than 500 core confor mations even for quite large and flexible ligands and far fewer for more rigid ligands Every rotamer state for each rotamer group is enumerated and the core plus all possible rotamer group conformations is docked as a single object in Glide Because each coretypically has many rotamer group combinations the effective number of conformations being docked can easily number in the thousands or tens of thousands for molecules having several rotatable bonds A key issue is how dosely one of the conformations matches the correct cocrystallized conformation An exact match is not needed because the ligand subse quently undergoes flexible torsional op
148. mental binding affinities are better than 10 uM An ensemble of such compounds can therefore be used to optimize the scoring function Because of the wide range of novel terms that have been incorporated it has been necessary to perform optimizations using a wide variety of receptors and active compounds The data set used to date is far from complete in covering the range of potential binding motifs but is significantly larger and more diverse than any previous data set used in the literature for this purpose The optimization process consists of adjusting parameters so that as few database ligands as possible achieve better scores than the tight binding better than 10 uM known actives for the term or terms under consideration In practice it is not possible to achieve perfect rank ordering in this regard for the reasons discussed previously The average error in binding affinity prediction in the PDB data set is about 1 7 kcal mol for properly docked ligands with some outliers with errors of more than 3 kcal mol These errors may be somewhat larger when cross docking rather than self docking is performed On the basis of this analysis some database ligands with experi mental binding affinities in the 10 100 uM range are likely to be computed as having low micromolar or even nanomolar binding affinity while at the same time some of the active compounds will be underpredicted by similar amounts As a result overpredicted database lig
149. ments emulate to some extent the effect of breathing motions a protein site might make to accommodate a tight binding ligand that is slightly larger than the native cocrystallized ligand Cross docking tests have consistentl y shown that it is important to modify the final vdW surface in this J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1757 manner Too much scaling however is detrimental be cause active ligands may no longer make suitably spe cific interactions with the receptor if the cavity is too large Moreover ligands that are too large to bind to the physical receptor may begin to dock and score well computationally swelling the ranks of the false posi tives The object is to find a happy medium between too little scaling and too much Table 3 shows how the choice of scaling factors effects the enrichment obtained in the database screens de scribed in section 3 Thermolysin is an example of a rigid open site while the p38 MAP kinasesite is highly mobile and the estrogen receptor contains a tightly enclosed hydrophobic channel Alternative scalings are shown for cases in which the preferred scaling previ ously found for Glide 1 8 or 2 0 differs from the current default scaling The table shows that the new default scaling works as well as or better than the previously identified preferential scaling in six of the eight cases The original scaling gives substantially better enrich ment factors for lerr and perform
150. method in which multiple structures are employed in docking and or induced fit methods are utilized to directly incorporate protein flexibility is facilitated The parameterization of XP Glide is carried out using a large and diverse training set comprising 15 different receptors and between 4 and 106 well docked ligands per receptor A separately developed test set incorporating four new receptors and additional ligands for two receptors already in the training set is also defined AII of the receptor and ligand data is publicly available as is our decoy set which has been posted on the Schrodinger Web site and is freely available for downloading and we provide extensive references documenting the origin of each ligand The results reported below have been obtained with the Glide 4 0 release The development of data sets suitable for the analysis described above is highly labor intensive consequently our current test set is too small to draw robust conclusions and the results reported herein must be regarded as preliminary While the test set results are encouraging with regard to demonstration of a respectable degree of transferability a rigorous assessment of the performance to be expected on a novel receptor will have to be performed in future publications Nevertheless qualitative and consistent improvement in the results for both training and test set at least as compared to the alternative scoring functions available in Glide is
151. ms button menu choose 5 Only the ligand and the nearby residues are displayed All residues that do not have any atoms within 5 of the ligand are undisplayed Hiding the residues that do not come into contact with the ligand makes it easier to examine the ligand receptor interactions You can also open the Atom Selection dialog box from the Display only button menu to pick the ligand and add atoms within a given radius of a particular set of atoms This approach allows you more flexibility in picking the atoms Glide 5 5 Quick Start Guide 31 Chapter 4 Examining Glide Data 32 4 4 Visualizing Hydrogen Bonds and Contacts In this exercise you will display hydrogen bonds between the ligand and the receptor To display hydrogen bonds between any two sets of atoms use the Atom set 1 and Atom set 2 selection options in the H Bonds tab 1 In the Project Table choose Entry gt View Poses gt Display H bonds Hydrogen bonds to the currently displayed pose are displayed as yellow dashed lines If you want to change the cutoffs for defining hydrogen bonds choose Entry gt View Poses Define H bonds and change the values in the Measurements panel which is dis played by choosing this item If you want a count of hydrogen bonds between the ligands and the receptor choose Entry View Poses Count H bonds An HBond property will be added to the Project Table and the count may take a few seconds to finish 2 Use
152. ms deviations given by Glide and GOLD and by Glide and FlexX for common sets of noncovalently bound ligands having up to 10 and up to 20 rotatable bonds as well as for all ligands Glide can handle i e up to 35 rotatable bonds In some cases the PDB structure available when the GOLD or FlexX work was done is no longer accessible but a later structure is available In such cases we have used the later submission For example Glide uses 4aah whereas GOLD and FlexX use 3aah The Glide calculations use the conformationally optimized versions of the native ligands The comparison for ligands having 10 or fewer rotatable bonds seems to us the most relevant to database screening applications which usually seek to find leads that are relatively inflexible On average Glide gives rms deviations that are less than 6096 of those given by GOLD and less than half those given by FlexX The comparison for ligands having up to 20 rotatable bonds is also favorable Table 5 presents the detailed results on which the summaries in Tables 3 and 4 are based This listing shows that Glide gives a better result nearly twice as often as GOLD and more than 4 times as often as FlexX Whilefar from perfect we believethat this performance represents a qualitative improvement in docking ac curacy We should note however that these compari sons may not be completely fair to GOLD and FlexX Theligands were prepared in a comparable manner for J ournal of Medici
153. n 1ett HIV protease Ihpx and factor Xa 1fjs Active ligands in this case are invariably large and fill multiple pockets in the active site HIV protease actives make a substantial number of hydrogen bonds and derive their binding affinity from this and the hydrophobic pair term Remarkably there is no contribution from the hydrophobic enclosure Thrombin ligands typically form a salt bridge with a buried Asp 189 carboxylate in one of the small available pockets and form other hydrogen bonds as well Hydrophobic enclosure also makes no contribution for thrombin ligands Factor Xa exhibits the same salt bridge motif but also has an unusual hydrophobic location in which a ring moiety of the ligand is sandwiched between a number of aromatic rings in a location near the surface of the protein This pocket provides some hydrophobic enclosure although not to the same degree as a deeply buried pocket like that in 1kv2 lcx2 or Irtl and can also accommodate pi cation and pi stacking interactions Discriminating false positives that display interactions with the protein surface rather than in binding pockets is important for these receptors particularly factor Xa 6 Small hydrophilic sites in which strong electrostatic interactions are important This category includes the neurami didase 1bji and GIuR2 receptors In neuramididase much of the binding affinity derives from salt bridges and the various ligands serve as important test cases for t
154. nal Chemistry 2004 Vol 47 No 7 1743 all three methods i e by subjecting the native ligand structures to a force field based opti mization procedure before docking though Glide also used conforma tional search to ensure that the starting ligand struc tures had no memory of the cocrystallized pose A differencearises in the preparation of the protein sites however in that both the GOLD and FlexX calculations retained the original PDB coordinates for non hydrogen atoms It is possiblethat these methods might give more accurate dockings if they used our protein and ligand preparations in which steric clashes have been an nealed away However whilethe GOLD study cites four cases of noncovalently bound ligands in which the crystallographic ligand geometry appears to be incorrect lapt 1tdb Ihef and 1ive it does not ascribe any of the problematic dockings to steric clashes between the ligand and the protein for FlexX only one such instance 1srj is cited 8 One reason for suspecting that GOLD and FlexX may beless sensiti veto the details of the protein preparation is that neither uses the hard 12 6 Lennard J ones vdW potenti al employed by Glide GOL D does employ an 8 4 potential but this potential is much more forgiving of nonbonded incursions For example it penalizes a contact at 7096 of the sum of the vdW radii by only 0 9 1 8 kcal mol whereas Glide penalizes such a contact by 5 5 11 kcal mol assuming a well
155. nds all ligands 72 cases 86 cases 93 cases av max av max av max method rmsd rmsd rmsd rmsd rmsd rms Glide 1 46 8 5 1 65 8 5 185 137 GOLD 2 56 14 0 2 92 14 0 3 06 14 0 Table 4 Comparison of rms Deviations for Flexible Docking by Glide and FlexX lt 10 rotable bonds lt 20 rotatable bonds all ligands 133 cases 175 cases 189 cases av max av max av max method rmsd rmsd rmsd rmsd rmsd rms Glide 1 38 8 5 1 70 9 2 195 137 FlexX 2 99 12 6 3 48 13 4 3 72 15 5 Both however are quite modest for sets of ligands having 0 8 or 0 10 rotatable bonds such as are often employed in database screens carried out to find new leads In general the docking performance of Glide is very reasonable over a wide range of rotatable bonds and chemical functionality Detailed results are given in Table S1 Supporting Information these results show that Glide reproduces the experimentally mea sured binding affinities for 128 cocrystallized ligands with an rms deviation of 2 3 kcal mol and produces rms deviations from the cocrystallized ligand position that are less than 1 for nearly half of the 282 cases and aregreater than 2 in only about one third of the cases Comparison to GOLD and FlexX Detailed docking accuracy results for GOLD and FlexX are posted on the GOLDP and FlexX web sites These data have enabled us to make the head to head comparisons shown in the tables below Tables 3 and 4 compare r
156. ne kinase ligands to the protein from the 1kim complex For database ligands they used 990 randomly chosen compounds from a filtered version of the ACD database Only the 1kim ligand dT and one other ligand idu are reported to be submicromolar binders The others five are also pyrimidine derivatives and three are purines acv gcv pcv range in activity from 1 5 to 200 uM Realistically speaking computational screening of compound databases usually can only hope to discover micromolar ligands This poses a stiff challenge because Charifson et al 1 found that the docking methods they surveyed performed reasonably well at finding low nanomolar binders seeded into a database screen but fell off rapidly in efficacy as the activity of the known binders decreased The ability to identify micromolar binders in a database screen is therefore a stringent and relevant test Figures 2a and 3 examine the thymidine kinase screen In this case Figure 3a uses the seven pyrimidine and three purine based ligands defined by Rognan and co workers as known actives while Figure 3b uses only the seven pyrimidines Comparison of the lowest bar segments shows that Glide 2 5 is significantly more effective than Glide 1 8 or 2 0 at concentrating known Enrichment Factors in Database Screening 100 BO 60 a b o P i n 20 30 50 100 1 2 3 5710 2030 50 100 100 lt T Z pci 80 d i 60 gt 1err 40K d 1
157. ne farnesyltransferase inhibitors design of macrocyclic compounds with improved pharmacoki netics and excellent cell potency J Med Chem 2002 45 2388 2409 23 The reduction in the vdW radii is needed to roughly preserve hydrogen bonding distances which reflect a balance between attractive electrostatic and repulsive vdW forces 24 Research Collaboratory for Structural Bioinformatics Rutgers University New Brunswick NJ http www rcsb org 25 Halgren T A MMFF VI MMFF 94s option for energy minimiza tion studies J Comput Chem 1999 20 720 729 26 For 1cps we used the native ligand structure because the ligand contains functionality that cannot be treated by MMFF94s The remaining four cases havelarger rings that Glide s conformation generator cannot handle and for which the conformational search yielded a different ring conformation For 1hbv and 1pro the MMFF94s optimized native ligand geometry was used instead while for 1d8f and 2hct conformational search was employed but the ring conformation was not sampled 27 Gasteiger J Rudolph C Sadowski J Automated generation of 3D atomic coordinates for organic molecules Terahedron Comput M ehodol 1990 3 537 547 28 Kramer B Rarey M Lengauer T Evaluation of the FlexX incremental construction algorithm for protein ligand docking Proteins 1999 37 228 241 29 The results for the FlexX test set are available at http cartan gmd
158. ng work with XP Glide is how to properly model binding when it is mainly hydrophobic in character These new develop ments will be described in a subsequent paper The most challenging problem in the use of docking methods in pharmaceutical applications is dealing with protein flexibility When the protein structures differ by discrete localized changes protonation state modi fications alterations of side chain rotamer states it should be possible to examine the variations in protein structure directly in a single docking run with suitable algorithms thus saving considerable computational effort When there is a larger perturbation of protein structure however as in cases such as 1kv2 in which there is a significant induced fit component of ligand binding that results in substantial backbone or loop movement other approaches are needed The simplest approach is to carry out flexible docking into multiple rigid protein structures and then to combine the screen ing results When knowledge of the appropriate en semble of protein structures is available this strategy is likely to succeed Examples of this approach will be described in a subsequent paper in the context of E xtra Precision Glide Ultimately accurate molecular mechanics modeling of the protein structure will be needed to enumerate the variations in active site geometry that can be accessed at relatively low energies Calculations along these lines if successful would
159. nged and 0 8 ligand scaling the same scalings were used to assess docking accuracy In the fifth section we compare these results with results obtained using scaling factors identified in earlier docking studies with Glide as giving optimal results The comparisons show that default scaling performs well though optimizing the scaling factors can improve the performance in some cases Measures of Virtual Screening Effectiveness To quantify Glide s ability to assign high ranks to ligands with known binding affinity we report enrichment factors in graphical and tabular form and present accumulation curves that show how the fraction of actives recovered varies with the percent of the database screened Following Pearlman and Charifson the enrichment factor can be written as EN H itSsamplea N sampled H itStotai N totai 1 J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1751 Equivalently this can be written as EF N totar N sampled H ItSsamplea H itStotat 2 Thus if only 10 of the scored and ranked database i e Ntotai N sampled 10 needs to be assayed to recover all of the Hitstota actives the enrichment factor would be 10 But if only half of the total number of known actives are found in this first 10 i e if Hitssampied Hitstota 0 5 the effective enrichment factor would be 5 Equation 2 is useful when sampling an initial small fraction of a database but to measure performance for recovering a
160. nt specificity pocket as displayed in Figure 8 c Hydrogen bonds of one ligand group to two different protein groups This requires having two like charged protein groups in close proximity This structure which presumably creates strain energy in the apo protein occurs with a greater frequency than might be expected Figure 9 presents an example showing ligand Gr217029 binding to the tern N9 influenza virus of the neuramidase receptor 1bji with a distance between carboxylate oxygen atoms of only 4 5 A Empirical observations such as the unexpectedly high potency of several neuramidase ligands including Gr217029 cited above and physical chemical reasoning in that the electric field from the two nonsalt bridged proximate carboxylates is highly negative and interacts more favorably with a ligand positive charge than is typical for a salt bridge suggest that c provides less stabilization energy than b which in turn provides less stabilization energy than a Similarly one would expect that a bidentate structure is more favorable electrostatically than a monodentate structure Note however that unless consider ation 1 is properly satisfied none of the three structures is likely to be favorable from a free energy point of view It is the Friesner et al Figure 8 The bidentate hydrogen bonds in this thrombin complex bridge the ligand and Asp 189 Figure 9 Gr217029 forms hydrogen bonds with two nearby Asp residues when
161. nt to other commercially available methods these results suggest that Glide 2 5 may represent a qualitative advance in scoring accuracy and virtual screening efficiency Comparisons made to GOLD 1 1 FlexX 1 8 and DOCK 4 01 for the thymidine kinase and estrogen receptors using datasets prepared by Rognan and co workers support this view though we again caution that these comparisons may not be representative of the current capabilities of these methods Glide 2 5 has a number of advantages relativeto pre vious versions One is that generally good results are obtained with the new default 1 0 protein O 8 ligand scaling Calibrating the scale factors can lead to im proved performance but this may be less critical than with earlier versions of Glide which employed less well balanced scoring functions A second advantage is that hydrogen bond filters i e imposition of a cutoff on the hydrogen bond energy and or metal ligation filters are nolonger necessary These elements broaden the range of applicability of Glide and simplify its use Onethemethat runs consistentl y through the results is that Glide does best when the active ligands make multiple hydrogen bonds tothe receptor and does worst when the site is hydrophobic and offers few such opportunities From what we have seen in theliterature this behavior is not unique to Glide One of the key problems in database screening one on which we have made considerable progress in ongoi
162. ntial Energy Mapping and Docking Studies Bioorg Med Chem 2006 14 3160 3173 Eldridge M D Murray C W Auton T R Paolinine G V Mee R P Empirical Scoring Functions I The Development of a Fast Empirical Scoring Function to Estimate the Binding Affinity of Ligands in Receptor Complexes J Comput Aid Mol Des 1997 11 425 445 Gehlhaar D K Verkhivker G M Rejto P A Sherman C J Fogel L J Freer S T Molecular Recognition of the Inhibitor AG 1343 by HIV 1 Protease Conformationally Flexible Docking by Evolutionary Programming Chem Biol 1995 2 317 324 10 Morris G M Goodsell D S Halliday R S Huey R Hart W E Belew R K Olsen A J Automated Docking Using a Lamarckian Genetic Algorithm and an Empirical Binding Free Energy Function J Comput Chem 1998 19 1639 1662 11 Muegge I Martin Y C A General and Fast Scoring Function for Protein Ligand Interactions A Simplified Potential Approach J Med Chem 1999 42 791 804 12 Bohm H J The Development of a Simple Empirical Scoring Function to Estimate the Binding Constant for a Protein Ligand Complex of Known 3 Dimensional Structure J Comput Aided Mol Des 1994 8 243 256 13 Bohm H J Prediction of Binding Constants of Protein Ligands A Fast Method for the Prioritization of Hits Obtained from De Novo Design or 3D Database Search Programs J Comput Aided Mol Des 1998 12 3
163. obviate the need for cocrystallized examples While modeling of this type is clearly quite difficult at present methods using continuum solvati on models such as those developed by Schr dinger in principle can address this problem effectively If this can be accomplished it would greatl y enhance the effecti ve ness of any docking methodology in a wide range of practical applications Efforts along these lines are Enrichment Factors in Database Screening currently underway at Schr dinger and have yielded promising early results Acknowledgment This work was supported in part by grants to R A F from the NIH Grants P41 RR06892 and GM 52018 We thank Dr Didier Rognan for providing electronic copies of receptor ligand and decoy data sets for the thymidine kinase and estrogen recep tors and of therank orders found for GOLD FlexX and DOCK Supporting Information Available Tables S1 S15 listing ranks of the active ligands their GlideScore values their hydrogen bonding scores and their Coluomb vdW in teraction energies with the protein site for the 15 database screens considered in this paper This material is available free of charge via the Internet at http pubs acs org References 1 Friesner R A Banks J L Murphy R B Halgren T A Klicic J J Mainz D T Repasky M P Knoll E H Shelley M Perry J K Shaw D E Francis P Shenkin P S Glide A new approach for rapid accurate
164. ocked and scored and the receptor and ligand poses are examined in Chapter 4 Protein preparation is not covered in this manual see the Protein Preparation Guide for details of this task Panel specific online help is available for all Glide panels If you need help with a Glide task click the Help button or see the Glide User Manual To complete the exercises you must have access to an installed version of Maestro 9 0 and Glide 5 5 For installation instructions see the nstallation Guide Exercises in some chapters produce structure files that are needed in subsequent exercises To allow you to begin at any exercise you choose these and other necessary files ligand files for example are included with the Glide distribution 1 1 About Glide and Maestro Glide is designed to assist you in high throughput screening of potential ligands based on binding mode and affinity for a given receptor molecule You can compare ligand scores with those of other test ligands or compare ligand geometries with those of a reference ligand Additionally you can use Glide to generate one or more plausible binding modes for a newly designed ligand Once you locate favorable structures or bonding conformations with Glide you can use Liaison or QSite to obtain binding energies for ligand receptor pairs Protein Preparation is usually required for Glide calculations It can be performed for most protein and protein ligand complex PDB structures using the Pro
165. olecule is challenging with respect to making additional hydrogen bonds such as those found in the bulk environment Our previous analysis of hydrophobic interactions suggests that the environment will be significantly more challenging if the water molecule has hydrophobic protein atoms on two faces as opposed to a single face and if few neighboring waters are available to readjust themselves to the constrained geometry of the protein water hydrogen bond Geometries of this type are identified using a modified version of the hydrophobic enclosure detection algorithm described previously Replacement of such water molecules by the ligand will be particularly favorable if the donor or acceptor atom of the ligand achieves its full complement of hydrogen bonds by making the single targeted hydrogen bond with the protein group in question so that satisfaction of additional hydrogen bonds is not an issue An example of a suitable group would be a planar nitrogen in an aromatic ring binding for example to a protein N H backbone group This has been observed to be essential to achieving high potency experimentally in the 1bI7 ligand binding to p38 MAP kinase as shown in Figure 3 Here the Met 109 hydrogen bond is known to be important for potency Analogous hydrogen bonds have been found to be important in other kinases In the absence of rigorous physical chemical simulations we have used the experimental data from a significant number of diverse
166. omponent is the introduction of a solvation model Like other scoring functions of this type previous versions of GlideScore did not properly take into account the severe restrictions on possible ligand poses that arise from the requirement that charged and polar groups of both the ligand and protein be adequately solvated Charged groups in particular require very careful assessment of their access to solvent In addition water molecules may be trapped in hydrophobic pockets by the ligand also an unfavor able situation To indude solvation effects Glide 2 5 docks explicit waters into the binding site for each energetically 1742 J ournal of Medicinal Chemistry 2004 Vol 47 No 7 competitive ligand pose and employs empirical scoring terms that measure the exposure of various groups to the explicit waters This water scoring technology has been made efficient by the use of grid based algorithms Using explicit waters as opposed to a continuum solvation model has significant advantages In the highly constrained environment of a protein active site containing a bound ligand the location and environment of individual water molecules become important Cur rent continuum solvation models have difficulty captur ing these details but our explicit water approach has allowed us to develop consistently reliable descriptors for rejecting a high fraction of the false positives that appear in any empirical docking calculation Our analy
167. onated flurbiprofen ligand 33 ML 3000 ligand 11 and deprotonated ML 3000 ligand 31 Examina tion of various combinations of protein and ligand scaling factors using Glide 1 8 led us to select 1 0 protei n O 8 ligand scaling which we also used here and for Glide 2 0 These Glide dockings have one unusual feature namely that only 23 of the 33 known actives dock with negative Coulomb vdW energies even with relatively heavy scaling of protein and ligand nonpolar vdW radii when a normally sized docking box centered around the 1cx2 ligand is used When the docking box is made much larger the remaining 10 ligands dock success fully but occupy a site that is displaced by 10 12 from the primary site However the site 2 ligands have relatively poor Coulomb vdW interaction energies and often make no hydrogen bonds It thus seems unlikely that these Cox 2 ligands actually dock into this second site The most likely explanation is that some variable element in the site geometry is responsible and that the limitations of docking to a rigid site are particularly extreme in this case Figures 2m and 9b show that Glide 2 5 performs much better than Glide 1 8 or 2 0 It does particularly well for early enrichment as indicated by the lowest seg ments of the bar chart in thelatter figure Indeed Glide 2 5 places 9 of the actives in the first 20 positions in the ranked database Table 13S Supporting I nforma tion Figure 9b and
168. only the single strongest interaction when the ligation is bi or multidentate A third is that Glide 2 5 reduces the net ionic charges for most charged charged and charged polar interactions but leaves them unchanged for metal ligand interac tions Thus the Coulombic contribution to GlideScore 2 5 uses the full Zn ligand interaction energy further helping to differentiate anionic ligands These elements were also included in GlideScore 2 0 but GlideScore 2 5 goes one step further by recognizing that neutral ligands such as imidazoles can be effective binders when Zn and the tripod of protein residues on which it sits are net neutral eg when Zn is coordinated by two glutamates and a histidine as in farnesyl protein transferase rather than by one glutamate and two histidines as in thermolysin In such cases the term in GlideScore 2 5 that rewards a geometrically appropriate metal ligation by 2 0 kcal mol when the ligand functionality is anionic is omitted when the apo site is net neutral While this modification seems appropriate further studies will be needed to determine whether it gets the balance right for net neutral sites 8 Cox 2 1cx2 We obtained structures for 33 known binders from the literature These include the native 1cx2 ligand SC 558 ligand 24 celecoxib ligand 2 rofecoxib ligand 3 indomethacin ligand 10 deproto nated indomethacin ligand 26 flurbiprofen ligand 25 deprot
169. ons with specific protein ligand interactions above and beyond the generic terms that have appeared repeatedly in prior scoring functions We have found that a relatively small number of such motifs are dominant over a wide range of test cases the ability to automatically recognize these motifs and assign binding affinity contributions potentially represents an advance in the modeling of protein ligand interactions based on an empirical scheme In section 3 we evaluate the performance of our methodology in self docking with regard to both the ability to generate the correct binding mode of the complex and the prediction of binding affinity using docked XP structures for the complexes In section 4 the performance of the scoring function in enrichment studies ability to rank known active compounds ahead of random database ligands for a substantial number of targets containing qualitatively different types of active sites is investigated Our treatment of the data differs significantly from what has generally prevailed in previous papers in the literature in evaluating scoring accuracy we distinguish cases where there are significant errors in structural prediction as opposed to systems where the structural prediction is reasonably good but the scoring function fails to assign the appropriate binding affinity By using only well docked structures to parametrize and assess scoring functions a way forward toward a globally accurate
170. or for rigid docking each ligand Glide performs an exhaustive search of possible positions and orientations over the active site of the protein The search begins with the selection of site points on an equally spaced 2 grid that permeates the active site region step 1 in Figure 1 To make this selection precomputed distances from the site point to the recep tor surface evaluated at a series of prespecified direc tions and binned in 1 ranges are compared to binned distances from the ligand center to the ligand surface For flexible docking the ligand center is defined as the midpoint of the two most widely separated atoms in the core region which includes the directly attached atom of each rotamer group for rigid docking the two most widely separated atoms in the entire ligand are used The line through these atoms is called the ligand diameter Glide positions the ligand center at the site point if there is a good enough match of the histograms of binned distances but skips over the site point if there is not Theligand center can be placed at any site point on or within a box by default 12 A on a side that contains the candidate site points The generous size of this ligand center box ensures that the placement of the docked ligand is not overly constrained The second stage examines the placement of atoms that lie within a specified distance of the ligand diam eter axis for a prespecified selection of possible orien
171. ost list table The number of processors on each host as given in the hosts file is shown in the Proces sors column 4 Edit the value in the Use column for the selected host to specify the number of processors to use The value in the Total to use text box is updated to reflect the value you entered If you want to use multiple hosts you can select multiple table rows and specify the number of processors to use on each The total number is given in the Total to use text box Glide 5 5 Quick Start Guide 20 Chapter 3 Ligand Docking A Ligand Docking Start x Output Incorporate Do not incorporate Job Standard names glide dock SP 1 Name glide dock SP 1 Compose Username dyall Separate job into 1 381 subjobs Host list faith 1 1 1 levanna T 1 ginger 1 1 iman 2 2 marci 1 Total to use EE fq w E Reset All IE Start Cancel Help Figure 3 4 The Start dialog box default settings The 10 subjobs will be distributed over the number of processors you specify We suggest you use five processors Then the first five subjobs will be sent first followed by each of the remaining five as processors become available When you are satisfied with your distributed processing settings continue to Section 3 5 3 5 Starting the Ligand Docking Job 1 If you did not do so in the previo
172. ow markers 5 Click the Fit to screen toolbar button The view zooms in so that the ligand fills most of the Workspace 6 Click in an empty portion of the Workspace to clear the selection 7 Next turn on Workspace Feedback by going to Display gt Show Single Entry Feedback or by typing S By default Workspace Feedback appears in the upper right corner of the Workspace and includes the entry Title GlideScore and EmodelScore Glide 5 5 Quick Start Guide Chapter 4 Examining Glide Data Now you are ready to view the poses 8 Press the RIGHT ARROW key The second pose replaces the first in the Workspace The RIGHT ARROW and LEFT ARROW keys can be used to step through the selected poses 9 Shift click the entry for the first pose The first pose is added to the Workspace 10 Press the RIGHT ARROW key The third pose replaces the first two in the Workspace 11 Reselect the first pose by clicking its In column 4 3 Displaying Atoms by Proximity In this section you will select a display that includes the ligand and the receptor residues nearest the ligand This is useful for examining contacts and hydrogen bonds between the ligand and the active site of the receptor From the Display only selected atoms button menu choose Molecules it 2 Click on an atom in the ligand The ligand molecule is displayed and all other atoms are undisplayed 3 From the Display residues within N of currently displayed ato
173. pounds We have carried out some preliminary investiga tions data not shown that suggest qualitatively reasonable results can be obtained in a limited number of test cases for ranking active compounds when cross docking is required but these results are not yet robust across a wide range of receptors and the quantitative precision with which this can be done is not yet clear 4 Enrichment Studies The goal of the present work is to optimize a scoring function that will properly assign binding affinities to active compounds if the compound is docked in a sufficiently native like pose and that will minimize the number of inactive database ligands that score well The question of what is sufficiently native like is a heuristic one but we would argue that precision in drawing the line is not critical If a few additional compounds are included or excluded this will have minimal effect on the overall optimization process Therefore visual inspection has been employed to construct data sets of active compounds that fit into the particular versions of the receptors employed Typically 60 100 of the available active compounds per receptor fell into this category so the current protocol is not simply cherry picking of a small number of compounds On the other hand the fact that poorly fitting compounds are not included in the data set must be taken into account when comparing with other results reported in the literature We have divided
174. pounds is to be robustly separated from a random database without false positives or false negatives significant work needs to be done to better treat receptor flexibility and ensure reliable docking accuracy However we believe that the attempt in this Table 14 Standard Enrichment Factors for Recovering 40 of the Correctly Docked Active Ligands in the Test Set enrichment factors screen v4 0 XP v4 0 SP BACE 30 25 factor VIIa 19 15 PPARY closed form 9 19 PPARy open form 14 40 Vegfr2 closed form 25 10 Vegfr2 open form 11 2 human cyclin dep kinase 8 3 thrombin 64 25 a Active ligands have at least 10 uM activity Table 15 Standard Enrichment Factors for Recovering 40 and 70 of Known Active Ligands in the Test Set Including Misdocked Cases enrichment factors v4 0 XP v4 0 SP screen 4096 70 40 70 BACE 12 3 11 0 factor VIIa 10 6 7 5 PPAR closed form 3 2 7 4 PPARy open form 3 3 11 7 Vegfr2 closed form 2 1 1 1 Vegfr2 open form 3 2 1 1 human cyclin dep kinase 4 3 2 2 thrombin 19 3 11 6 paper to separate scoring function accuracy docking accuracy and reorganization energy effects is an essential starting point if truly robust approaches are to be developed 5 Conclusions We have described a novel scoring function and an enhanced sampling algorithm the combination of which constitutes the Glide 4 0 XP docking methodology The methodology has been tested with a diverse se
175. pportunities cf the hy drogen bondi ng scores in Tables S14 and S15 Support ing Information It therefore qualifies as a difficult site for an empirical scoring function such as GlideScore 2 5 In such a site it is crucial to recognize and penalize mismatches in complementarity in the docked poses The improved performance relative to the earlier ver sions of Glide reflects the better balance of the more widely parametrized GlideScore 2 5 function as well as the inclusion of desolvation penalties these terms play an even larger role in the Extra Precision Glide 2 5 scoring function 3 4 Comparison to Other Methods Comparisons usually are difficult for us to make because we do not have access to other docking codes and because published comparisons often use propri etary datasets 111419 n this section however we present comparisons to published results for GOLD 1 1 3 F lexx 1 8 4 and DOCK 4 0129 for the thymidine kinase and estrogen receptors using datasets provided to us by D Rognan We caution that the earlier versions of GOLD FlexX and DOCK used by Rognan and co workers may not be representative of the current capabilities of these methods However the same comparisons were also used by J ain in his recent paper introducing the Surflex method Their study used GOLD FlexX and DOCK as docking engines and employed ChemScore FlexX the DOCK energy score GOLD PMF F resno and Score to rank the docked pos
176. protein required to accom modate the ligand In many cases the protein is relatively rigid the ligand fits without major rearrangements and neglect of this term is acceptable However there are cases where it is overwhelmingly likely that the induced fit energies are sub stantial The most obvious cases are those in which an allosteric pocket is created to accommodate the ligand This occurs for example in nonnucleoside reverse transcriptase inhibitors NNRTIs of HIV reverse transcriptase HIV RT and is also manifested in the large scale motion of the activation loop in p38 MAP kinases required to produce the DFG out conforma tion to which inhibitors such as BIRB796 bind However there are many more subtle cases in which side chains or backbone groups alter their positions nontrivially and this should introduce some energetic cost The goal of an empirical scoring function should be to predict the binding affinity of the ligand to the structure with which it is presented That is given the formal impossibility of predicting reorganization energy in any such scheme this value should be removed from the experimental binding affinity Otherwise one will be fitting to an incorrect experimental number given the goal of the exercise A perfect scoring function of this type will correctly rank order candidate ligands in their ability to bind to the structure at hand If this structure is known to have a low reorganization energy compounds
177. r Grid Generation panel showing the H bond Metal subtab 2 In the Constraints tab click on the H bond Metal tab 3 Ensure that Pick atoms is selected 4 Click on the carboxyl oxygen that is hydrogen bonded to the amidine of the ligand This atom is the OD1 atom of ASP 189 When you have picked the atom an entry is added to the Receptor atoms table The Name column shows the identity of the atom with hbond in parentheses In the Atom column both oxygen atoms of the carboxylate are listed in square brackets because Glide includes symmetry related atoms as part of the constraint Both atoms are marked in the Workspace with a padlock icon if Show mark ers is selected and the name is also displayed if Label atoms is selected Glide 5 5 Quick Start Guide Chapter 2 Receptor Grid Generation 5 Name the constraint S1 site by editing the text in the Name column 6 From the Display H bonds button menu choose Delete H bonds k H 2 5 Starting and Monitoring the Grid Calculation With the ligand and the active site defined and constraints set up the grid generation job can be started 1 In the Receptor Grid Generation panel click Start You might see a dialog box that warns about an unidentified ligand This is a check for ligand sized molecules that might be in the active site It has found the glycerine mole cules which are not a problem for docking You can therefore ignore the warning and click Continue
178. r RMSD avg RMSD avg all ligands with AGexp 198 2 26 1 75 3 18 2 51 all well docked ligands 136 1 73 1 34 2 80 2 23 all poorly docked ligands 62 3 02 2 49 3 70 3 11 RMS and average absolute deviations are presented in kcal mol Well docked ligands are defined as those having an RMSD to the cocrystallized pose of 2 5 or less plus those identified in Table 5 as being docked appropriately but in a symmetry related orientation Table 7 Training Set Used to Characterize XP Virtual Screening PDB no no well docked code description actives actives 1e66 acetylcholinesterase 20 20 1bji neuraminidase 9 9 Ifjs factor Xa 13 8 lkv2 human p38 map kinase 10 10 1b17 p38 map kinase 36 21 Irt1 HIV RT 29 23 1cx2 cyclooxygenase 2 13 13 lagl human cyclin dep kinase 10 6 lett thrombin 16 15 Ihpx HIV 1 protease 14 9 3ert human estrogen receptor 10 8 lqpe Ick kinase 121 87 Im17 EGRE tyrosine kinase 117 106 1tmn thermolysin 6 3 kim thymidine kinase 4 4 2 All correctly docked ligands have experimental activities lt 10 uM except those for neuraminidase hydrophobic enclosure reorganization energy by capping the maximum assignable enclosure energy at a value of 2 0 kcal mol as compared to the maximum scoring function value of 4 5 kcal mol It also appears as though the special bidentate charged charged reward when the ligand group is positively charged involves some degree of protein reorganization pre sumably to enable the bidentat
179. r described above assigns scores to lipophilic ligand atoms based on summation over a pair function each term of which depends on the interatomic distance between a ligand atom and a neighboring lipophilic protein atom This clearly captures a significant component of the physics of the hydrophobic component of ligand binding It is assumed that Journal of Medicinal Chemistry 2006 Vol 49 No 21 6179 A B Figure 1 Schematic of a ligand group interacting with two distinct hydrophobic environments above a hydrophobic plane A and enclosed in a hydrophobic cavity B the displacement of water molecules from areas with many proximal lipophilic protein atoms will result in lower free energy than displacement from areas with fewer such atoms As a crude example it is clear that if the ligand is placed in an active site cavity as opposed to on the surface of the protein the lipophilic atoms of the ligand are likely to receive better scores If they are located in a hydrophobic pocket of the protein scores should be better than in a location surrounded primarily by polar or charged groups Furthermore these improved scores are likely to be correlated with improvements in ligand binding affinity However a function dependent only on the sum of interatomic pair functions is potentially inadequately sensitive to details of the local geometry of the lipophilic protein atoms relative to the ligand lipophilic atom in question A
180. radii we recommend for use when the protein site has not been relaxed to remove possible steric dashes see section 5 We also used docking and scoring grids of the same size employed in section 3 As previously noted Rognan and co workers randomly selected the 990 database ligands from a filtered version of the ACD database They employed various scoring functions in conjunction with each of the docking methods but we focus here on the native GOLD docking GOL D scoring F lexX docking F exX scor ing and DOCK docking DOCK scoring combinations These are the ones most likely to be used in a pharma ceutical setting where project needs may not permit extensive explorations of alternative docking scoring combinations to be carried out Comparisons of docking results for Glide GOLD FlexX and DOCK are presented in Table 2 and in Figures 11 and 12 These comparisons show that DOCK docking followed by DOCK energy scoring is the worst model Glide 2 5 appears to be the best model overall by virtue of its superior performance for thymidine kinase though Surflex does even better for this recep tor and though Glide 1 8 and 2 0 do better for the estrogen receptor cf Figure 12 and Table 2 The latter may reflect the better balance of the Glide 2 5 scoring funcion which often yields better results for poorly treated screens at the cost of some degradation in performance for well handled screens FlexX and GOLD show decent enrichment but n
181. rating features that contribute to the specialized scoring function terms as appropriate Results of Training Set Enrichment Studies Table 9 reports a measure of enrichment defined as the average number of database ligands outranking the active compounds in the database Specifically the number of database ligands with a GlideScore that is superior to each active is tabulated these values are summed and the result is then divided by the total number of active compounds in the data set We believe that this metric is superior to standard definitions of enrichment which punish active ligands when they are outranked by other active ligands this is a particularly serious problem when the active test suite contains a large number of compounds A perfect score based on this metric would thus be zero no database ligands outranking any active compounds and smaller numbers are better These values are also presented for the older 2 7 XP and 4 0 SP Glide results As noted previously only active ligands that successfully docked in 4 0 XP Glide were consid ered In a small number of cases active ligands failed to dock with 4 0 SP or 2 7 XP For a calculation of the number of outranking decoy ligands such ligands were ranked lower than all successfully docked active and decoy ligands Table 10 is the corresponding table constructed using a more standard definition of enrichment that we have employed 6194 Journal of Medicinal Chemistry 2006 Vo
182. receptor plus ligand complex The script also sets thetautomeric statefor His residues which are assumed to be neutral by considering potential metal ligation and hydrogen bonding interactions Thethird step is to postprocess the pprep d receptor This is necessary because the judgments made by the preparation procedure will not always be correct For example for an aspartyl protease such as HIV both active site aspartates will be dose to one or more atoms of a properly docked ligand and hence will be repre sented as negatively charged One of these aspartates however is typically taken to be neutral in modeling studies Similarly Glu 143 in thermolysin typically interacts with an acceptor oxygen of the bound ligand This residue may need to be protonated as may His 231 which forms a salt bridge with Asp 226 and typically Friesner amp al places its second ring nitrogen atom within about 3 of a second zinc bound oxygen of a carboxylate or phosphonate ligand Other special circumstances can also arise In addition if the protein had some or all hydrogens attached the original and prepared versions of the protein need to be compared to decide how to resolve any discrepancies Step four adds structural waters if any are to be kept Nearly all waters have been removed in this work Exceptions are ladd ladf lebg three waters 1lna two waters and 1mdr where a water molecule is tightly bound to a Mg ion By comparison G
183. rk New York 10036 Schr dinger L L C 120 W 45th Street New York New York 10036 Schr dinger L L C 1500 SW First Avenue Portland Oregon 97201 and D E Shaw Research and Devdopment 120 W 45th Street New York New York 10036 Received December 24 2003 Unlike other methods for docking ligands tothe rigid 3D structure of a known protein receptor Glide approximates a complete systematic search of the conformational orientational and positional space of the docked ligand In this search an initial rough positioning and scoring phase that dramatically narrows the search space is followed by torsionally flexible energy optimization on an OPLS AA nonbonded potential grid for a few hundred surviving candidate poses The very best candidates are further refined via a Monte Carlo sampling of pose conformation in some cases this is crucial to obtaining an accurate docked pose Selection of the best docked pose uses a model energy function that combines empirical and force field based terms Docking accuracy is assessed by redocking ligands from 282 cocrystallized PDB complexes starting from conformationally optimized ligand geometries that bear no memory of the correctly docked pose Errors in geometry for the top ranked pose are less than 1 in nearly half of the cases and are greater than 2 in only about one third of them Comparisons to published data on rms deviations show that Glide is nearly twice as accurate as GOLD and more than
184. s make an unusual number of hydrophobic contacts A penalty is assigned if the number of such contacts for an individual water molecule exceeds a given threshold Water scoring statistics are also used to determine whether special hydrogen bonding rewards should be assigned as was discussed previ ously Contact Penalties Ei strain Penalizing strain energy in rigid receptor docking is probably the single most difficult component of an empirical scoring function The problem arises from the fact that in a typical cross docking situation the ligand has to adjust to fit into an imperfect from its point of view and rigid cavity This often requires ligands to adopt higher energy nonideal torsion angles Considering the rigid receptor approximation that is made it is difficult to determine whether strained ligand geometries would arise if induced fit effects were properly accounted for or whether strained ligand geometries would be a true requirement for docking to that receptor Furthermore even native ligand geometries have been found to exhibit high strain energies Given this limitation the function used to penalize poses with close internal contacts is fairly lenient and only looks for severe cases of bad internal 6186 Journal of Medicinal Chemistry 2006 Vol 49 No 21 contacts One function simply counts the number of intramo lecular heavy atom contacts below roughly 2 2 and rejects a pose entirely if there are more than
185. s or to the interaction between or interoperability of Schr dinger products and services and such other third party software June 2009 Contents Document Conventions nee v Chapter T Getting Sale een 1 1 1 About Glide and Maestro sse 1 1 2 Preparing a Working Directory sse 2 1 3 Starting Maestro and Setting the Working Directory 3 Chapter 2 Receptor Grid Generation ss 5 2 1 Importing the Prepared Structures sss 5 2 2 Defiring tlie Recepto crisan O mern 6 2 3 Defining the Active Site ssssssssssseserserrrsnenrerensnnenernerrenenrenen rn renen arenan anar annen anar nan ann 7 2 4 Setting Up Glide Constraints mmmnmnssssrsresrssessrssesssenrssr enes saras area aan ananas nare 9 2 4 1 Setting the Display for Constraint Definition 9 2 4 2 Defining a Positional ConstralDi uiri rte tr rrr RH PH HERE RIpRES 10 2 4 3 Defining an H bond Coristrailil 25 ccrto nennen 11 2 5 Starting and Monitoring the Grid Calculation sess 13 Chapter 9 Ligand Docking 15 3 1 Specifying a Set of Grid Files and Basic Options ee 15 3 2 Specifying Ligands To Dock sss 17 3 3 Specifying Output Quantity and File Type sse 18 3 4 Setting Up Distributed Processing sss 19 3 5 Starting the Ligan
186. s an example consider the two model distributions shown in Figure 1 In one case A a lipophilic ligand group is placed at a hydrophobic wall with lipophilic protein atoms on only a single face of the hydrophobic group In the second case B the lipophilic ligand group is placed into a tight pocket with lipophilic protein atoms contacting the two faces of the ligand group As suggested above one would normally expect a larger contribution to binding in the second case than in the first However this does not fully settle the question which at root is whether the atom atom pair contribution for a given ligand atom protein atom distance should be identical when the ligand atom is enclosed by protein hydrophobic atoms as opposed to when it is not or whether there can be expected to be nonadditive effects From a rigorous point of view the answer depends principally upon the free energy to be gained by displacing a water molecule at a given location This in turn depends on how successfully that water molecule is able to satisfy its hydrogen bonding requirements at that location while retaining orientational flexibility In the extreme case in which a single water molecule is placed in a protein cavity that can accommodate only one water molecule and is surrounded on all sides by lipophilic atoms that cannot make hydrogen bonds the enthalpy gain of transferring the water to bulk solution is enormously favorable In such a case it is
187. s presented in Table 11 enable a direct connection to be made with alternative viewpoints In what follows our discussion is focused on the results in Table 9 for the reasons given above The XP 4 0 results are nearly uniformly comparable to or better than those of either SP 4 0 or XP 2 7 and in many cases are significantly better as is manifested with particular clarity using the new definition of enrichment There is a slight degradation for the estrogen receptor from XP 4 0 for cyclooxy genase 2 relative to both XP 2 7 and SP 4 0 but all of the results for these test cases are very good The real question with regard to scoring function effectiveness is the ability to prevent false positives from ranking ahead of active compounds XP 4 0 displays an ability to reduce the average number of false positives ranking ahead of actives in many cases by an order of magnitude and in some cases by nearly 2 orders of magnitude as compared to both 2 7 XP and 4 0 SP This same effect is also reflected in the more common definition of enrichment factor Table 10 but the improvement is quantitatively obscured by the definition of enrichment employed particularly for the data sets containing larger numbers of actives For example in EFGR kinase the number of actives is greater than 1046 of the random database and standard enrichment measures that effectively penalize active compounds for having other active compounds ranking ahead of other actives can
188. s somewhat better for 1tmn but the enrichment factors are high in these cases and the default enrichments are also good Table 4 shows that 0 9 0 8 scaling occasionally allows one or two additional actives to dock Moreover the more generous scaling almost always produces a sig nificantly lower rank for thelast common active found e g the 8th for 3ert the 9th for lerr or the 21st for 1cx2 This too indicates that 0 9 0 8 scaling produces a better physical model when the fit is tight The 1rt1 screen is an exception because 0 9 0 8 is not in fact the optimal scaling for Glide 2 5 The condusion we draw is that use of optimal scaling factors should be considered for high performance screens When active ligands are unavailable or will not be used to determine the scaling factors the current default should normally be used H owever if the protein heavy atom coordinates are taken directly from the X ray structure it may be better to use 0 9 0 8 scaling to reduce the effect of unresolved steric clashes This more generous scaling should also be used in cases in which it is known that the active site region is tight and enclosed an example being the hydrophobic channel of the estrogen receptor because it will be difficult in such cases for certain active ligands to avoid serious steric clashes with the rigid site Conversely a lesser degree of scaling might be tried if the site is open and is known to be relatively rigid 6 Dis
189. scoring functions appropriate protein and ligand prepara tion is particularly critical In practical applications it is often necessary to carry out such preparation without prior knowledge of the binding mode of the complex However for the present purposes the objective is to optimize and evaluate the scoring function for correctly docked compounds Therefore we have endeavored to use all available information in preparing ligand and protein structures The most problematic aspect of protein and ligand preparation is the assignment of protonation states of ligand and protein in the protein active site note that we neutralize ionizable residues distant from the active site to mimic the effects of dielectric screening by solvent and counterions In the absence of structural data correctly making such assignments can be a very challenging task However when structural data is available the most likely protonation states of both protein and ligand can usually be deduced from the structure of the complex In cases for which we do not have a PDB structure the correct binding mode of the ligand can typically be inferred by analogy With a binding mode and solution phase pK s of the ligand the correct protonation states can then be assigned It should be emphasized that the results shown here require accurate protonation state assignment and substantial degradation can result from incorrect assignments in unfavorable cases A second aspect of
190. side chain at a time thereby avoiding the combinatorial explosion of total molecular conformations that occurs when all side chains are considered together Because the anchor fragment is already positioned in the protein most side chain conformations can be trivially rejected based on steric clashes The Glide rough scoring function is used to screen the initial side chain conformations Most importantly these conformations can be grown at ex tremely high resolution 4 degrees for each rotatable bond because the total number of conformations considered at any one time is being constantly pruned via screening and clustering algorithms It is this high resolution sampling that enables Friesner et al difficult cross docking cases to be effectively addressed and that ultimately allows penalties to be avoided when possible After the individual side chains are grown a set of candidate complete molecules is selected by combining high scoring individual conformations at each position and eliminating structures with significant steric clashes between side chains Candidate structures are minimized using the standard Glide total energy function which employs a distance dependent dielectric to screen electrostatic interactions and are ranked according to the Emodel Glide pose selection function com prising the molecular mechanics energy plus empirical scoring terms Then the grid based water addition technology is applied to a subset of th
191. sition dialog box opens In the Select atoms to define a position section ensure that Pick is selected and from the Pick option menu that Atoms is selected Click the carbon atom that is between the two nitrogen atoms in the imidazole ring in the ligand A semi transparent gray sphere is displayed around the atom Enter the name S4_arom in the Name text box Enter 2 0 in the Radius text box Click OK The constraint is added to the Positions table in the Positional tab and the sphere changes to yellow The name is displayed next to the sphere _ New Position Select atoms to define a position ASL atom num 27 X All Selectior Previous Select Pick Atoms Show markers Name S4 arom Radius 1 0 OK Cancel Figure 2 4 The New Position dialog box Glide 5 5 Quick Start Guide Chapter 2 Receptor Grid Generation Receptor Grid Generation EES Receptor Site Constraints Rotatable Groups 1 constraints have been defined limit is 10 total Positional 1 H bond Metal 0 Hydrophobic 0 Define the positions of spherical regions that should be occupied by particular ligand atoms during docking The position of each sphere is the centroid of the picked atoms Occupation of these spheres may be chosen as constraints during docking Positions S4 arom 7 36 2 54 18 21 1 00 New Delete
192. solvation energy terms protein ligand structural motifs leading to enhanced binding affinity are included 1 hydrophobic enclosure where groups of lipophilic ligand atoms are enclosed on opposite faces by lipophilic protein atoms 2 neutral neutral single or correlated hydrogen bonds in a hydrophobically enclosed environment and 3 five categories of charged charged hydrogen bonds The XP scoring function and docking protocol have been developed to reproduce experimental binding affinities for a set of 198 complexes RMSDs of 2 26 and 1 73 kcal mol over all and well docked ligands respectively and to yield quality enrichments for a set of fifteen screens of pharmaceutical importance Enrichment results demonstrate the importance of the novel XP molecular recognition and water scoring in separating active and inactive ligands and avoiding false positives 1 Introduction In two previous papers we have described the Glide high throughput docking program and provided performance bench marks for docking and scoring capabilities These results have established Glide as a competitive methodology in both areas 2 gt However it is clear from enrichment results ref 2 that there remains substantial room for improvement in separating active from inactive compounds In this paper we outline and present results obtained from significantly enhanced sampling methods and scoring functions hereafter collectively referred to as extra
193. sonable root mean square deviation RMSD as compared to the native complex or as obtained by analogy with the native complex of a related ligand Comparison by analogy is often necessary when dealing with a large dataset of active ligands only a few of which may have available crystal structures Our discussion of XP Glide is divided into four different sections First in section 2 we describe the novel terms leading to enhanced binding affinity that have been introduced to account for our observations with regard to protein ligand binding in a wide range of systems The origin of these terms lies in the theoretical physical chemistry of protein ligand interactions however developing heuristic mathematical rep resentations that can be used effectively in an empirical scoring function taking into account imperfections in structures due to the rigid receptor approximation and or limitations of the docking algorithm requires extensive analysis of and fitting 2006 American Chemical Society Published on Web 09 23 2006 6178 Journal of Medicinal Chemistry 2006 Vol 49 No 21 to experimental data Key aspects of this analysis along with illustrative examples are provided in section 2 in an effort to provide physical insight as well as formal justification for the model In developing XP Glide we have attempted to identify the principal driving forces and structural motifs for achieving significant binding affinity contributi
194. ssed Phone 212 854 7606 Fax 212 854 7454 E mail rich chem columbia edu t Columbia University Schr dinger L L C NY Present address Serono International S A CH 1211 Geneva 20 Switzerland Schr dinger L L C OR Schr dinger L L C NY and D E Shaw Research and Develop ment feasible while retaining sufficient computational speed to screen large libraries This has been accomplished via the use of a series of hierarchical filters as described below The current performance characteristics of Glide are as follows i Docking times average less than 1 min for data sets having 0 10 rotatable bonds on an AMD Athelon MP 1800 processor running Linux ii Robustness in binding mode prediction is quali tatively superior to what is reported in the current literature for docking methods in widespread use For example a comparison with results obtained by the developers of GOLD yields an average rmsd of 1 46 for Glide compared with 2 56 for GOLD for the 72 noncovalently bound cocrystallized ligands of the GOLD test set that have 10 or fewer rotatable bonds The comparison to FlexX is even more favorable Compari sons for ligands having up to 20 rotatable bonds yield similar results iii Binding affinity predictions compared with ex perimental data for cocrystallized complexes are rea sonable 2 3 kcal mol rmsd though dearly subject to improvement iv Results for library screening reported
195. sting grid files during a docking job Bounding Box The green cube shaped marker that appears in the Workspace during Glide docking job setup after you select active site residues coordinates or a ligand to be used as the box s center The box represents the space in which ligands are allowed to move during docking Increasing the size of the bounding box increases the space that can be sampled by the docked ligands and consequently increases the CPU time required for the calculation Contacts Graphical representations of the van der Waals interactions between the atoms of two or more molecules Within Maestro contacts are categorized as Good Bad and Ugly Good contacts are those that have van der Waals radii consistent with the experimen tally determined values for the involved atom types Bad contacts depict those interactions that are experimentally improbable Ugly contacts represent van der Waals interactions that are disallowed in experimental systems Enclosing Box The purple cube shaped marker that appears in the Workspace after you specify active residue sites coordinates or a ligand to be used as a bounding box center using the Glide panel The enclosing box represents the space that any part of any specified ligand can sample during a docking calculation Compare this with the green bounding box which represents the space that the center of each specified ligand must be confined to during a docking calcu
196. substantial fraction of the active ligands we prefer to modify the definition of enrichment as follows EF 50 WAPR ai Hits suy Hitsqu 3 In this equation APRsampied is the average percentile rank of the Hitssampea Known actives Intuitively this makes sense if the actives are uniformly distributed over the entire ranked database the average percentile rank for an active would be 50 and the enrichment factor would be 1 Unlike egs 1 and 2 however this formula considers the rank of each of the Hitssampied known actives not just the rank of thelast active found which is what Nsampiea iS likely to be As a result the enrichment factor will be larger than the value com puted from eq 1 or 2 if the actives are concentrated toward the beginning of the Nsampied ranked positions but will smaller if the actives are grouped toward the end of this list This is appropriate because a key objective in database screening is to find active com pounds as early as possiblein the ranked database the new definition is better at indicating when this is happening 3 Virtual Screening Results Overview of Glide s Performance in Database Screening Table 1 compares the performance of Glide 1 8 2 0 and 2 5 SP standard precision using as definitions of enrichment EF 70 which measures the enrichment for recovering 70 of the known actives and EF 296 which measures enrichment for assaying the top 296 of the ranked database These r
197. t match select All Glide 5 5 Quick Start Guide Chapter 3 Ligand Docking 7 Edit Feature Feature Donor I5 Import Export Pattern list Dp 2717 U 1 7 1 m 1 S X2 1 0 1 1 0 X2 1 Edit Delete Apply marker offset Help Figure 3 6 The Edit Feature dialog box For H bond and hydrophobic constraints the ligand features that must match these constraints are predefined You can edit them if you want but this is not necessary For positional constraints you must define the ligand feature that matches the constraint Features are defined in terms of SMARTS patterns 4 From the Available constraints table select the S4 arom row and click Edit Feature The Edit Feature dialog box opens There are no SMARTS patterns in the Pattern list table because the Custom feature type is undefined by default Click New The New Pattern dialog box opens Enter the following text into the SMARTS pattern text box a This pattern matches all aromatic atoms Enter 1 into the Numbers text box This is the index of the atom in the SMARTS pattern that is matched by the constraint Since there is only one atom in the pattern this is the only possible choice Click OK The New Pattern dialog box closes and a row is added to the Pattern list table in the Edit Feature dialog box Glide 5 5 Q
198. t of ligands and receptors and has produced large improvements in binding affinity prediction and database enrichment as compared to other scoring functions within Glide The potential for providing physical insight into the origins of enhanced binding affinity is in our view as important as quantitative improvement of enrichment factors Visualization of XP Glide terms as is presented in the Figures of the present paper can be utilized by modelers and medicinal chemists in the design of new inhibitors The success of design efforts along these lines in the context of lead optimization would provide 6196 Journal of Medicinal Chemistry 2006 Vol 49 No 21 the most convincing evidence that the underlying model of molecular recognition proposed herein has substantial validity Our hope is that the present paper will facilitate work along these lines by describing in considerable detail the theory that underlies the XP Glide implementation Acknowledgment We thank Mark Murcko Bob Pearlstein and Barry Honig for reading preliminary versions of this manuscript and providing useful feedback We also thank Mike Campbell for assistance in generating graphics Supporting Information Available Detailed descriptions of the algorithms used in hydrophobic enclosure scoring and in scaling special neutral neutral hydrogen bond motif rewards References to experimental binding affinities for all test and training set ligands are also included T
199. ta tions of the ligand diameter step 2a If there are too many steric clashes with the receptor the orientation is skipped Next step 2b the ligand is rotated about the ligand diameter and the subset consisting of the atoms capable of making hydrogen bonds or ligand metal interactions with the receptor is scored subset test If this scoreis good enough all interactions with the receptor are scored step 2c The scoring in these three tests is carried out using a discretized version of ChemScore in which precom puted scores for the ChemScore atom types are assigned to 1 boxes Much as for ChemScore itself this algo rithm recognizes favorable hydrophobic hydrogen bond ing and metal ligation interactions and also penalizes steric clashes This stage is called greedy scoring be cause the actual score for each atom depends not only on its position relative to the receptor but also on the best possible score it could get by moving 1 in X Y and or Z This is done to mute the sting of the large 2 jumps in thesite point ligand center positions The final step in stage 2 isto rescore the top greedy scoring poses typically 5000 in number via a refinement proce dure step 2d in which theligand as a wholeis allowed to move rigidly by 1 A in the Cartesian directions Energy Minimization Using a Molecular Me chanics Scoring Function Only a small number of the best refined poses typically 400 are minimized on
200. tagonists as well as agonists Our studies used the 10 low nanomolar ERa antagonists that Rognan selected as active binders This is one casein which the nonbonded radii need to be scaled down to allow the known binders to dock correctly F or example five of the known binders had positive Coulomb vdW interaction energies when no scaling was done For Glide 1 8 and 2 0 weoriginally used 0 9 protein O 8 ligand scaling but we employ the default 1 0 0 8 scaling here Figures 2b c and 4 show that both estrogen receptor sites aretreated very well by all three scoring functions Tables S2 and S3 Supporting I nformation list the Glide 2 5 rankings 1754 J ournal of Medicinal Chemistry 2004 Vol 47 No 7 a 18 20 25 18 20 25 Figure 5 Percent of CDK 2 kinase actives recovered for assaying the top 296 596 and 1096 of the ranked database a 1dm2 site b laql site 100 80 0 0 1 8 2 0 2 5 1 8 2 0 2 5 0 1 82 02 5 Figure 6 Percent of p38 MAP kinase actives recovered for assaying the top 296 596 and 1096 of the ranked database a 1a9u site b 1bl7 site c 1kv2 site 3 CDK 2 Kinase 1dm2 1aq1 The CDK 2 kinase site is highly flexible A key issue is the length of the rather narrow binding cavity which if insufficient will prevent many active ligands from correctly docking After examining superimposed structures for five co crystallized PDB complexes we chose a site from the 1dm2 compl
201. te The reason is that the GIn 125 side chain undergoes a 180 rotation on going from a pyrimidinesiteto a purine site and the geometry that is correct for the parent site has an acceptor acceptor and or a donor donor dash in the alternative site For theseven pyrimidine based ligands Glide does very well except for hmtt which does not fit quite properly into the 1kim site when the nonpolar ligand vdW radii are scaled by 0 8 the default Surflex and GOLD also give 1744 journal of Medicinal Chemistry 2004 Vol 47 No 7 Friesner amp al Table 5 The rms Deviations for Glide GOLD and FlexX for Members of the GOLD and FlexX Test Sets complex Glide GOLD FlexX complex Glide GOLD Flexx complex Glide GOLD FlexX complex Glide GOLD FlexX 121p 157 n a 1 29 laaq 130 1285 175 labe 0 17 0 86 116 labf 0 20 n a 1 27 lag 0 28 400 049 lam 029 0 81 1 39 laco 1 02 0 86 0 96 laha 011 051 0 56 lake 3 35 n a 1 18 lapt 058 1 62 1 89 latl 0 94 n a 2 06 lavd 052 n a 1 22 lazm 187 252 2 37 lbaf 076 612 827 1bbp 4 96 n a 3 75 lbma 9 31 n a 13 41 lbyb 10 49 n a 1 62 Icbs 196 n a 1 68 1cbx 0 36 054 1 35 1cde 129 n a 7 45 1cdg 3 98 n a 4 87 1cil 3 82 n a 3 85 1com 3 64 n a 1 62 1coy 028 086 1 06 Icps 3 00 0 84 0 99 Ictr 3 56 n a 2 82 Idbb 0 41 117 O81 1dbj 020 072 1 22 ldbk 0 47 n a 0 76 1dbm 1 97 n a 2 08 1did 3 88 3 72 4422 Idie 0 79 1 03 4 71 1dr1 147 141 5 64 ldwb 025 n a 0 54 Idwc 0 87 n a 1 19 ldwd 132 171 1 66 leap 2 330 3 00 372
202. ted number of compounds will be considered experimentally and each computa tionally identified compound needs to be as high in quality as possible In what follows we discuss the development and parametrization of Glide 2 5 SP XP docking and scoring will be described in a subsequent paper GlideScore 2 5 modifies and extends the ChemScore function as follows AGying Ciipo tipo f ri ag Chbond neut neut Z SAN h Aa Chbond neut chargea Z Ar h Aa zd Choond chargsa carad J DAT h Aa Coena fn Ei C H rotb C polar phobV polar phob xb Caoul E coul CyawE vaw solvation terms 2 The lipophilic lipophilic term is defined as in Chem Score The hydrogen bonding term also uses the Chem Score form but is separated into differently weighted components that depend on whether the donor and acceptor are both neutral one is neutral and the other is charged or both are charged In the optimized scoring function the first of these contributions is found to be the most stabilizing and the last the charged charged term is the least important The metal ligand interac tion term the fifth term in eq 2 uses the same functional form as is employed in ChemScore but varies in three principal ways First this term considers only interactions with anionic acceptor atoms such as either of the two oxygens of a carboxylate group This modi fication allows Glide to recognize the evident strong preference for coordination o
203. tein Preparation Wizard panel in Maestro For detailed information see the Protein Preparation Guide Maestro is Schr dinger s powerful unified multi platform graphical user interface GUI It is designed to simplify modeling tasks such as molecule building and data analysis and also to facilitate the set up and submission of jobs to Schr dinger s computational programs The main Maestro features include a project based data management facility a scripting language for automating large or repetitive tasks a wide range of useful display options a comprehen Glide 5 5 Quick Start Guide Chapter 1 Getting Started sive molecular builder and surfacing and entry plotting facilities For more detailed informa tion about the Maestro interface see the Maestro Overview the Maestro online help or the Maestro User Manual Maestro comes with automatic context sensitive help Auto Help Balloon Help tooltips an online help facility and a user manual For more information on getting help see page 35 You can also undo some operations in Maestro For more information see page 31 of the Maestro Overview The Impact computational engine is the underlying computational program for Glide It can perform molecular mechanics calculations either through the Maestro interface or from the command line For information on running basic Impact jobs see the Impact User Manual or the mpact Command Reference Manual 1 2 Preparing a Working
204. tems with exceptionally large protein reorganization energies as predicted by large hydrophobic enclosure contributions have been intentionally omitted This omission includes allosteric sites such as the HIV RT and p38 structures mentioned above and complexes such as staurospo rine CDK2 PDB code laql in which the CDK2 pocket must expand substantially to accommodate the unusually large staurosporine ligand If one believes that the scoring function is accurate the protein reorganization energy can be inferred from the computed rigid receptor and experimental binding 6188 Journal of Medicinal Chemistry 2006 Vol 49 No 21 Friesner et al Table 5 4 0 XP and SP Binding Scores and Heavy Atom RMS A Values for Docking into PDB Entries GlideScore kcal mol RMS GlideScore kcal mol RMS PDB AGexp XP XP corr SP XP SP PDB AGexp XP XP corr SP XP SP laaq 11 5 10 6 10 6 11 5 2 01 1 40 le5i 7 4 7 4 11 2 0 28 0 17 labe 8 9 1 8 1 8 8 8 0 31 0 40 leap 8 5 12 6 11 8 9 2 0 65 2 38 labf 7 4 8 0 8 0 9 3 0 17 0 14 lebg 14 8 11 0 11 0 18 3 0 34 0 26 lacj 10 0 11 4 9 9 8 4 2 81 4 61 lecv 6 6 7 4 7 4 9 4 0 24 0 18 lacm 10 3 12 2 122 13 1 0 40 0 32 leed 6 5 12 0 12 0 8 3 11 29 1 58 laco 4 9 2 3 2 3 10 1 0 34 0 29 lejn TI 9 8 7 8 10 0 0 34 0 12 ladd 9 2 10 3 8 3 9 9 0 83 0 70 le
205. the RIGHT ARROW and LEFT ARROW keys to step through the poses The hydrogen bonds are displayed as each pose is included in the Workspace Note the difference in hydrogen bonding patterns between the ligands In the pose list 344 Good vdW 6 Bad vdW and 1 Ugly vdW contacts are reported for 1dwd Even the least good ligand pose has many more good contacts than bad or ugly ones The default is to display only Bad or Ugly contacts between the ligand and the receptor To display contacts between any two sets of atoms use the Atom set 1 and Atom set 2 selection options in the Contacts tab 3 From the Entry menu choose View Poses Display Contacts The contacts are shown as dashed lines connecting Workspace atoms Ugly contacts are shown in red and Bad contacts are shown in orange By default atoms that are hydrogen bonded are not considered to have bad or ugly contacts If you want to change the cutoffs to redefine the distance criteria for Good Bad and Ugly contacts choose Entry gt View Poses gt Define Contacts and change the values in the Measurements panel which is displayed by choosing this item If you want a count of contacts between the ligands and the receptor choose Entry View Poses Count Contacts The count may take a few seconds to finish and three new prop erties are added to the Project Table Good Bad and Ugly 4 Use the RIGHT ARROW and LEFT ARROW keys to step through the poses Glide 5 5 Quick Start G
206. the Tyr 99 Trp 215 Phe 174 pocket Also a pi cation interaction is received by the charged end of the ligand Figure 12 The Ihpx ligand bound to HIV 1 protease Hydrophobic groups of the ligand are not hydrophobically enclosed and do not receive a hydrophobic enclosure packing reward For example the phenyl ring faces hydrophobic residues on only one face reorganization energy Figures 2 13 provide illustrative ex amples of a typical active ligand binding to the various Journal of Medicinal Chemistry 2006 Vol 49 No 21 6193 Figure 13 4 Hydroxytamoxifen bound to the human estrogen receptor Hydrophobic enclosure about the phenoxy group is illustrated by displaying lipophilic protein atoms as green spheres Table 9 Average Enrichments Defined as the Average Number of Outranking Decoy Ligands over Correctly Docked Actives in the Training Set avg number of outranking decoys screen v4 0 XP v2 7 XP v4 0 SP acetylcholinesterase 111 580 344 neuramididase 25 411 37 factor Xa 1 196 187 human p38 map kinase 26 30 57 p38 map kinase 7 93 183 HIV RT 11 26 83 cyclooxygenase 2 24 12 22 human cyclin dep kinase 3 TI 216 thrombin 2 57 70 HIV 1 protease 16 60 167 human estrogen receptor 14 2 23 Ick kinase 13 157 EGREF tyrosine kinase 41 411 279 thermolysin 9 107 32 thymidine kinase 0 1 29 2 Active ligands have at least 10 uM activity except those for neuramini dase as described in Section 4 receptors illust
207. the main window on the main toolbar click the Import structures button E The Import panel is displayed 2 From the Files of type menu ensure that Maestro is chosen Glide 5 5 Quick Start Guide Chapter 3 Ligand Docking 3 Click Options The Import Options dialog box opens 4 Ensure that Import all structures Replace Workspace and Fit to screen following import are all selected 5 From the Include in Workspace option menu ensure that First Imported Structure is cho sen 6 Click Close in the Import Options dialog box 7 Navigate to the structures subdirectory and select the file sar_reference mae gz 8 Click Open The reference ligand is displayed in the Workspace Next the settings for the previous constraints job need to be cleared 9 n the Ligand Docking panel click the Settings tab and select SP standard precision The other settings will be left as they were for the previous docking job 10 In the Constraints tab clear all check boxes in the Use column Receptor based constraints will no longer be applied instead you will be using ligand based constraints The core constraint is defined in the following steps 11 In the Core tab select Restrict docking to reference position The first few controls in the Define core section become available 12 Enter 1 5 in the Tolerance text box 13 Click an atom in the reference ligand in the Workspace The ligand is marked with purple markers and the re
208. the nth root of the product of then individual enrichment factors This definition was further modified by replacing any enrich ment factor of less than 1 by 1 and by weighting the contribu tions such that the composite weight is the same for each receptor type even when two or three receptor site geometries are used 6 S 7 9 gt J ournal of Medicinal Chemistry 2004 Vol 47 No 7 1759 11 Charifson P S Corkery J J Murcko M A Walters W P Consensus scoring A method of obtaining improved hit rates from docking databases of three dimensional structures into proteins J Med Chem 1999 42 5100 5109 12 J ain A N Surflex fully automatic flexible molecular docking using a molecular similarity based search engine J Med Chem 2003 46 499 511 13 Murphy R B Friesner R A Halgren T A Unpublished results 14 Stahl M Rarey M Detailed analysis of scoring functions for virtual screening J Med Chem 2001 44 1035 1042 15 Pargellis C Tong L Churchill L Cirillo P F Gilmore T Graham A G Grob P M Hickey E R Moss N Pav S Regan J Inhibition of p38 MAP kinase by utilizing a novel allosteric binding site Nat Struct Biol 2002 9 268 16 Whittaker M Floyd C D Brown P Gearing A J H Design and therapeutic application of matrix metalloproteinase inhibi tors Chem Rev 1999 99 2735 2776 17 Babine R E Bender
209. these is as follows 1 Small hydrophobic sites HIV RT 1rt1 and cyclooxy genase 2 Cox 2 1cx2 The HIV RT NNRTI site is an allosteric pocket that opens to accommodate the ligand the Cox 2 site does not display as dramatic a structural rearrangement but is also highly hydrophobic The dominant terms of the scoring function are the pair hydrophobic term and the hydrophobic enclosure term The hydrophobic enclosure is quite large in both cases for known active compounds In HIV RT the active compounds typically make a single hydrogen bond to a backbone carbonyl that receives the special single neutral neutral hydrogen bond reward discussed in section 2 In Cox 2 there is also typically a single hydrogen bond though it does not receive a special reward 2 Medium sized sites making a single special hydrogen bond This category includes EGFR tyrosine kinase and the 1bl7 form of p38 MAP kinase This motif is one of the two typical kinase binding motifs in which there is a special hydrogen bonding site in the hinge region of the kinase These systems allow only a single hydrogen bond in this site typically involving a ring nitrogen atom whereas other kinases form a correlated pair or triplet of hydrogen bonds discussed in 3 below 3 Sites making correlated hydrogen bonds This category includes thymidine kinase TK CDK2 and lck kinase LCK TK and LCK form a pair of correlated hydrogen bonds in the standard kinase hinge reg
210. timization We have found however that for best results one of the starting conformations needs to be within about 1 5 rmsd of the correct cocrystallized conformation We have tested our algorithm by applying it to 796 cocrystallized ligands taken from the PDB and by locating the core 1746 journal of Medicinal Chemistry 2004 Vol 47 No 7 Friesner amp al Table 9 The rms Deviations between Best Generated and Cocrystallized Ligands Expressed as a Percentage of Ligands Falling into Specified rms Distance Ranges percent of ligands having an rms deviation in the listed range no of rotatable bonds no of ligands 0 00 0 49 0 50 0 99 1 00 1 49 1 50 1 99 2 00 2 99 73 00 1 10 495 33 37 23 15 2 0 11 15 159 3 29 45 17 6 0 16 20 87 0 10 38 31 20 1 gt 20 55 0 9 25 29 27 9 plus rotamer group conformati on that best matches the cocrystallized pose The results summarized in Table 9 show that only 796 of ligands having 1 10 rotat able bonds have rms deviations between the best core rotamer group conformation and the cocrystallized pose of 1 5 or greater For more flexible ligands the errors areunderstandably larger Nevertheless 7796 of ligands having 11 15 rotatable bonds and 4896 of ligands having 16 20 rotatable bonds have rms devia tions of less than 1 5 Furthermore 89 of ligands with 11 20 rotatable bonds fall within 2 Initial Screening of Ligand Poses For each core conformation
211. tka ET f 11 5 2 28 2 28 5tmn 11 0 12 0 12 0 12 1 2 87 2 50 ltlp 10 3 10 0 10 0 10 4 7 70 7 33 6abp EST ZN 8 8 0 33 0 36 1tmn 10 0 10 2 10 2 10 4 3 65 1 90 6cpa 15 7 10 8 10 8 10 8 3 93 4 29 ltng 4 0 3 9 3 9 83 0 26 0 21 6rnt A E o 82 0 63 0 63 1tnh 4 6 3 9 3 9 8 7 0 39 0 28 6tim 4 4 5 9 5 9 8 3 0 59 0 42 Itni cS 3 7 3 7 6 7 2 12 2 03 6tmn 6 9 10 4 10 4 11 6 2 53 2 57 ltnj 2 7 4 1 4 1 7 8 0 43 0 36 7abp 8 6 8 0 8 0 9 4 0 15 0 17 1tnk 2 0 3 8 3 8 6 9 1 17 0 98 7cpa 19 0 13 1 13 1 10 5 3 91 2 87 1tnl 2 6 4 0 4 0 7 4 0 54 0 24 7cpp 5 2 6 2 6 2 1 99 3 23 Itph 3 1 9 9 2 3 7 5 0 22 0 23 7tim 74 5 4 54 7 7 0 20 0 19 Itpp full e 79 0 43 1 07 8abp 10 7 7 6 7 6 8 4 0 10 0 21 ltrk 9 6 9 6 11 2 1 63 2 15 8atc 10 3 9 5 9 5 10 7 0 41 0 38 Ityl 5 2 5 2 6 7 5 20 1 08 8gch 1 9 79 9 6 0 32 0 30 lukz 11 3 11 3 13 6 0 55 0 41 9abp 10 9 72 A 10 4 0 23 0 13 lulb 6 0 6 7 6 7 8 7 0 34 0 35 9hvp 11 4 11 9 11 9 12 9 1 44 1 47 RMS values are computed relative to the native cocrystallized ligand Scores are compared to available experimental binding energies AGexp in kcal mol Note that the XP scores have been adjusted to cap the hydrophobic enclosure packing term at a value of 2 0 to bring this term closer to an absolute
212. tment of Chemistry Columbia University New York New York 10036 and Schr dinger L L C 1500 SW First Avenue Portland Oregon 97201 Received December 24 2003 Glide s ability to identify active compounds in a database screen is characterized by applying Glide to a diverse set of nine protein receptors n many cases two or even three protein sites are employed to probe the sensitivity of the results to the site geometry To make the database screens as realistic as possible the screens use sets of druglike decoy ligands that have been selected to be representative of what we believe is likely to be found in the compound collection of a pharmaceutical or biotechnology company Results are presented for releases 1 8 2 0 and 2 5 of Glide The comparisons show that average measures for both early and global enrichment for Glide 2 5 are 3 times higher than for Glide 1 8 and more than 2 times higher than for Glide 2 0 because of better results for the least well handled screens This improvement in enrichment stems largely from the better balance of the more widely parametrized GlideScore 2 5 function and the inclusion of terms that penalize ligand protein interactions that violate established principles of physical chemistry particularly as it concerns the exposure to solvent of charged protein and ligand groups Comparisons to results for the thymidine kinase and estrogen receptors published by Rognan and co workers J Med
213. uick Start Guide 23 Chapter 3 Ligand Docking 24 7 New Pattern SMARTS pattern a Example C CH3 CH3 CH3 Get From Selection Numbers 1 OK Cancel Figure 3 7 The New Pattern dialog box 9 Click OK The Edit Feature dialog box closes This completes the definition of the Custom feature If you did not define this feature the docking job would not be started 10 Click Start The Start dialog box opens If you ran the SP docking job on multiple processors before you can do the same for this job 11 Select the host and number of processors 12 Change the job name to factorXa sp cons and click Start The Monitor panel opens and displays the progress of the job When the job finishes examine the results in the Project Table Of the 50 ligands poses are reported for only six ligands of which the first four are the actives and the other two did not score very well The remaining ligands did not satisfy the constraints These results indicate that the application of constraints serves to discriminate between ligands that bind in the proper mode and ligands that don t 3 8 Docking Ligands Using Core Constraints In this exercise you will apply core constraints defined by a pattern in the reference ligand and use these to dock a set of structures Structures that do not include the core pattern will not be docked The first task is to import the reference ligand 1 In
214. uide Chapter 4 Examining Glide Data 4 5 Visualizing Glide XP Descriptors In this exercise you will use the Glide XP Visualizer to examine the contributions of various terms to the XP scoring function The terms are given a spatial representation that you can display together with the ligand and the receptor Note This exercise uses the results obtained from the exercise in Section 3 9 on page 26 which requires a special license for XP descriptor generation Click the Clear Workspace toolbar button e 2 From the Applications menu choose Glide XP Visualizer The Glide XP Visualizer panel opens 3 Click Open 4 Select factorXa xp refine xpdes in the file selector that opens and click Open After a short delay the receptor and the highest scoring ligand are displayed in the Work space and the table in the panel is filled in The underlined values indicate that there is a corresponding visualization for this value 5 Click the PhobEn cell for ligand 16088 You might want to deselect Narrow columns to see the entire column heading This col umn displays the hydrophobic enclosure rewards After a few seconds the naphthalene of the ligand is displayed in ball and stick and the hydrophobic atoms on the protein sur rounding this ring are displayed in CPK in gray 6 Click the PhobEn cell for ligand 612278 Note that for this ligand there is only a benzene ring rather than a naphthalene 7 Click the HBond cell for
215. ult settings On a similar machine with a single processor the 50 ligand docking job will usually finish in about 45 minutes As 10 subjobs distributed over 5 similar processors the docking job will finish in about 10 minutes When the job finishes examine the results in the Project Table panel The four active ligands glide lignum 1 through 4 are ranked highest 3 6 Docking in High Throughput Virtual Screening Mode Glide has a set of predetermined options that can speed up the docking by a factor of about seven over the standard precision SP docking mode In this exercise you will run an HTVS docking job on the same set of ligands as used in the SP docking exercise 1 In the Settings tab select HTVS high throughput virtual screening The other settings will be left as they were for the SP docking job 2 In the Output tab ensure that Write ligand pose file excludes receptor filename will be lt jobname gt _lib mae is selected 3 Click Start The Start dialog box opens You should not need to run this job on multiple processors 4 Change the job name to factorXa_htvs 5 Select a host and set the number of processors and subjobs to 1 6 Click Start The job should take only a few minutes to run When the job finishes examine the results in the Project Table panel The four active ligands are ranked in the top 5 ligands Their scores differ a little from those in the SP docking run Glide 5 5 Quick Start Guide 21
216. ure of the receptor Having the receptor structure included in the file is convenient for displaying hydrogen bonds and contacts between the ligand and the receptor In Chapter 4 you will use tools in the Project Table to examine the poses in the file factorXa sp pv maegz Ligand Docking ER Settings Ligands Core Constraints Similarity Output Structure output Format Write pose viewer file includes receptor filename will be lt jobname gt _pv mae Write ligand pose file excludes receptor filename will be lt jobname gt _lib mae Write out at most 10000 ligand poses per docking run Write out at most poses per ligand Perform post docking minimization Number of poses per ligand to include 19 Threshold for rejecting minimized pose e s kcal mol CJ Apply strain correction terms Write per residue interaction scores for residues within 12 0 of grid center _ Compute RMSD to input ligand geometries _ Write report to file lt jobname gt rept start Write Reset Close Help Figure 3 3 The Output tab of the Ligand Docking panel Glide 5 5 Quick Start Guide Chapter 3 Ligand Docking 2 Ensure that the value of m in the Write out at most m poses per ligand text box is 1 the default Because there are only 50 ligands in the input file this setting ensures that no more than 50 poses one for each ligand will be
217. us section click Start The Start dialog box opens In the Start dialog box change the name to factorXa sp If you have not already done so select a host from the Host list na BB WwW N host enter the correct user name in the Username text box 6 Click Start The docking job starts and the Monitor panel is displayed Glide 5 5 Quick Start Guide From the Incorporate option menu choose Append new entries as a new group If you have a different user name on the host you have selected from that on your local Chapter 3 Ligand Docking For the distributed processing example as soon as the actorXa sp job has been launched it is divided into subjobs As each subjob is launched on a processor it is listed in the Monitor panel When one subjob is finished the next one is launched To view the log for any subjob select it in the job table and click Monitor If the subjob is already finished the entire log can be scrolled through in the File tab of the Monitor panel The results for each subjob are stored in subdirectories of the output directory and collected at the end into the output directory The time required for Glide docking jobs depends on the processor speed and workload the size and flexibility of the ligands and the volume specified by the enclosing box As a rough estimate docking a typical drug like ligand takes about one minute on a 1 2 to 1 5 GHz Pentium 4 processor under Linux using Glide 5 5 defa
218. used to quantitatively predict protein ligand binding in the highly heterogeneous and complex environment of a protein active site requires direct engagement with a critical mass of experimental data as well as extensive parameterization and investigation of a variety of specific functional forms In what follows we describe the results of our investigations along these lines A large number of computational experiments involving modifications of the hydrophobic scoring term designed to discriminate between different geometrical protein environments have been performed The criterion for success in these experiments is the ability of any proposed new term to fit a wide range of experimental binding free energy data and yield good predictions in enrichment studies Key findings are summarized as follows 1 Ligand hydrophobic atoms must be considered in groups as opposed to individually The free energy of water molecules in the protein cavity is adversely affected beyond the norm primarily when placed in an enclosed hydrophobic microenvi ronment that extends over the dimension of several atoms If there are individual isolated hydrophobic contacts the water will typically be able to make its complement of hydrogen bonds anyway by partnering with neighboring waters as in clathrate structures surrounding small hydrocarbons in water After empirical experimentation the minimum group size of connected ligand lipophilic atoms has been set at
219. verview of Docking Methodology Glide uses a series of hierarchical filters to search for possible locations of the ligand in the active site region of the receptor Figure 1 The shape and properties of the receptor are represented on a grid by different sets of fields that provide progressively more accurate scor ing of the ligand pose By pose we mean a complete spedification of the ligand position and orientation relativetothe receptor core conformation and rotamer group conformations These fields are generated as preprocessing steps in the calculation and hence need to be computed only once for each receptor The next step produces a set of initial ligand confor mations These conformations are selected from an exhaustive enumeration of the minima in the ligand torsion angle space and are represented in a compact combinatorial form Given these ligand conformations initial screens are performed over the entire phase space available to the ligand to locate promising ligand poses This prescreening drastically reduces the region of phase space over which computationally expensive energy and gradient evaluations will later be performed while at the same time avoiding the use of stochastic methods such methods can miss key phase space regions a certain fraction of the time thus precluding development of a truly robust algorithm To our knowl edge Glide is unique in its reliance on the techniques of exhaustive systematic search t
220. vities lt 10 uM When multiple forms of a receptor are utilized the number of actives reported here is all known actives with affinities lt 10 uM Ligands are assessed as optimally fitting into either the open or closed form of the receptor For example in PPARy 61 of known active ligands 57 of the 93 were assessed as fitting into either the open or the closed form of the receptor Table 13 Average Enrichments Defined as the Average Number of Outranking Decoy Ligands over Correctly Docked Actives in the Test Set avg no of outranking decoys screen v4 0 XP v4 0 SP BACE 35 342 factor VIIa 30 75 PPARy closed form 45 48 PPARy open form 44 76 Vegfr2 closed form 52 222 Vegfr2 open form 69 310 human cyclin dep kinase 25 206 thrombin 9 52 2 Active ligands have at least 10 uM activity is a case where SP scoring performs unexpectedly well as opposed to suggesting a particular problem with the XP scoring function A number of caveats should be emphasized with regard to these results The test set is small and there are almost certainly cases where enrichment performance will not be as good as that indicated in Tables 13 and 14 Furthermore high enrichment with few false positives can only be expected when the active ligands are properly docked The fact that a significant fraction of actives are not well docked even when two receptor conformations are used indicates that if a diverse set of active com
221. which hydrogen bonding is evalu ated While GlideScore 2 0 differentiated hydrogen bonds on the basis of charge our investigation of database screening results and PDB cocrystallized structures led us to a new understanding of how hydrogen bonds can be further differentiated We be lieve the contribution of hydrogen bonds to binding 1748 journal of Medicinal Chemistry 2004 Vol 47 No 7 affinity depends substantially on the details of the hydrogen bonding in certain specific ways On the basis of these insights we programmed terms beyond those shown in eq 2 into Glide and optimized them against our entire range of database screens Whilethis is still an ongoing research area the current parametrization leads to a substantial improvement compared to the treatment in Glide 2 0 One similarity however is that neutral neutral hydrogen bonds make the largest contribution followed by charged neutral and then charged charged hydrogen bonds Protein Preparation Our philosophy in the ap plication of rigid docking methods to virtual screening is that if possible information based on existing coc rystallized structures for the receptor of interest should be exploited Obviously there will be situations in which no experimentally determined structure exists e g when dealing with genomic targets In this case a variety of strategies are possible for example the use of homology modeling based on cocrystallized structures for relate
222. which the ligand charge is placed is sufficiently electrostatically favorable The electrostatic field at the ligand site is summed using constant and distance dependent dielectric models and cutoffs are imposed for assigning rewards based on empirical optimization over our suite of test cases These cutoffs help reduce the number of false positives receiving special charged charged rewards Table 3 enumerates the various special charged charged rewards for motifs based on the five categories discussed above The numerical values have been optimized based on fitting to our entire test suite Table 4 displays the XP active scores for a series of Glur2 receptors versus the experimental binding energies Along with neuramididase this is a system for which electrostatic interac tions are particularly important As such it provides an important contribution to the training set The good agreement displayed was achieved by using a combination of the electrostatic terms discussed above Other Terms A number of other types of specialized terms have been investigated These include terms rewarding pi stacking and pi cation interactions Epi rewards for halogen atoms placed in hydrophobic regions and an empirical correc tion enhancing the binding affinity of smaller ligands relative to larger ones These parameterizations were in many cases performed using limited data and we do not view them yet as fully mature As such details will not be pr
223. with good scores when docked into that structure should yield satisfactory experimental binding affinities For allosteric pockets and other sites with larger reorganization energies one would expect that more favorable scores would be needed to yield the desired experi mental binding affinity The problem of comparing scores between ligands docked into different conformations of the receptor can then be treated separately The problem is highly nontrivial requiring either a heuristic procedure incorporating experimental information or the brute force ability to compare free energies of different protein conformations A qualitative observation that we have made confirmed in a large number of examples is that a large hydrophobic enclosure score is a signature of significant protein rearrangement and possibly creation of an allosteric pocket Ligands binding tightly to both the HIV RT NNRTI site and p38 DFG out conformation for example generally receive maximal hydrophobic enclosure scores Furthermore the total XP scores of these ligands are substantially higher in absolute terms than their experimental binding affinities would mandate This is completely consistent with the ideas discussed above in which the reorganization energy of the protein must be subtracted from the empirical binding affinity score to produce the correct experimental binding affinity In the initial Glide XP parameterization with PDB cocrys tallized structures sys
224. yield enrichment factors of at most 10 3 for Table 10 and 9 5 for Table 11 The results shown in Table 9 are not perfect However until intrinsic RMS fluctuations in the scoring function can be reduced from the present average of 1 7 kcal mol for well docked ligands the scoring function seems unlikely to systematically perform significantly better without overfitting The number of high scoring database ligands reflected in this table is consistent with the estimated experimental population of low micromolar Friesner et al hits in a 1000 molecule random database of drug like molecules The acetylcholinesterase receptor appears to manifest the largest systematic errors This is likely due to our inability to optimize the pi cation and pi stacking scoring function terms with high precision because we lack sufficiently diverse examples mani festing these terms There also remain some difficulties associ ated with smaller highly hydrophobic sites such as Cox 2 and in medium sized sites with a single special hydrogen bond such as EGFR Overall though the results are reasonably robust across the entire data set and clearly represent a major advance over the results obtained using 4 0 SP or 2 7 XP Direct comparisons with other codes would require using the same sets of actives and database ligands Based on anecdotal reports from various sources and from comparison with published data Glide SP has generally performed at least as well in enri

Glide-related material 1. Richard A. Friesner, Jay L. Banks, Robert B

Contents

Download Pdf Manuals

Related Search

Related Contents