Home

Acuity 4.0 User Manual

Contents

1. 84 Web Links and Pathways 85 Reproducing the Published 86 UMM ALY tude E 86 Feedback 87 Technical ASsistancee cccssccsssssssscssscsscsccccscessesesssseneesecsscsssssseesesssseseesessssenees 89 Customer License Agreement cscssccsscsssssccsssssssccssseseesssssceseesesssceseesecsseees 91 Licensing NOUICOsiesicccsssscacsenessccesssdsasessecscenssseuetesonsieusecodenadeseveesessSsacecostesdeesdeasdeseve 93 Table of Contents Installation e 1 Chapter 1 Installation Computer Requirements Because Acuity is a client server application there will be slightly different requirements for client and server computers In general server computers that store data and run the database should have larger hard disks and faster hard disk access depending on the analyses being performed client computers need more RAM and faster processors Minimum Client or Server Requirements If you intend to run the Acuity client and server on the same computer the following is a minimum configuration IBM AT compatible computer with a Pentium 1 GHz or faster processor Windows 98 or ME operating system dual boot systems are not recommended 256 MB RAM Hard disk with 10 GB free for data storage CD ROM drive Chapter 1 2 Installation e 1024x768 display system with 65K colors e Internet Explorer 5 0
2. 2 38 Summar AON 40 Algorithm Complexity 41 LITA A E E tee eee 42 Chapter 4 Tutorials denssaotesees 47 Vint DUCE OM ER 47 Starting Acuity and Connecting To Database 48 Starting Acuity 48 Connecting To 48 Changing Your 49 Forgotten 49 Importing Microarray Data 49 The Acuity Interfaces 50 Common Tasks ies 50 Project eee nee Tee 51 Performing 63 64 Creating a 66 Preparing Dataset for Analysis cccccesccessceesceeseeeseeeseeescecsaecseenseenseenseenes 68 Finding Differentially Expressed Genes 71 Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Table of Contents v Hierarchical Clustering 78 Visualizing Principal Components Analysis SOMs Together
3. EAE A weet bases 10 Chapter Analysis Algorithms Theory and 11 The Fundamental Assumption 11 What Clustering 5 12 8 E E cect Paice 13 PCA r 14 Which Clustering Method Should I Use Data 14 1 5 3 15 Hierarchical Cluster 1 16 Correlation Coefficients cccccccccescessceseceescseeeeeeeeeseeeseeeseeeseecsaeeaecnseenseenaeens 17 Table of Contents iv Table of Contents K Means and K Medians Cluster Analysis 24 Gap Statistic Analysis 26 Self Organizing Maps Analysis 27 GeneShavihg aa E A A 29 Algorithm D etails 30 Principal Components Analysis 30 Normalization a a a 32 Origins of tae odes ees eee dike 32 Normalization 32 Linear Ratio 34 Lowess Normalization ccccccccssscssecssecesecesecesecseeeeeeeeeseceseeeeecsaecaecsseenaeenaeenas 35 Robust Multichip Analysis RMA 38 Background Correction 38
4. e Double click on a Self Organizing Maps analysis result to open it into the new window e Select Window Tile Horizontal to tile the windows Expand the displays by clicking on each analysis result in turn and selecting View Expand or use the Hot Key Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 85 You should now have one analysis in one window and second analysis in the second window e Select a Self Organizing Maps cluster and the substances are highlighted in the principal components analysis display e Select a region in the principal components display and the closest matching cluster is selected in the Self Organizing Maps display You can use the principal components analysis display to investigate any cluster solution by selecting the cluster and seeing the points that are selected in the principal components analysis display e Open a principal components analysis and a Self Organizing Maps or other cluster analysis as described above e Select any cluster from a hierarchical or non hierarchical method e substances are selected in the principal components analysis display e If they form a tight set with no unselected substances in their midst then the cluster forms a homogeneous group of substances If the cluster is mixed with unselected substances then it may not sufficiently distinct from other clusters to make it an interesti
5. Some states do not allow the exclusion or limitation of implied warranties or liability for incidental or consequential damages so the above limitations or exclusions may not apply to you U S Government Restricted Rights The SOFTWARE and its documentation are provided with RESTRICTED RIGHTS Use duplication or disclosure by the U S Government is subject to restrictions as set forth in subparagraph 1 11 of The Rights in Technical Data and Computer Software clause at DFARS 252 227 7013 or subparagraphs c 1 and 2 of the Commercial Computer Software Restricted Rights at 48 CFR 52 227 19 or clause 18 52 227 86 d of the NASA Supplement to the FAR as applicable Manufacturer is Molecular Devices Corp 1311 Orleans Drive Sunnyvale CA 94089 1136 USA Governing Body This Agreement is governed by the laws of the State of California Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Licensing Notice e 93 Licensing Notice Axon Instruments Molecular Devices is not licensed under any patents owned by Oxford Gene Technology Limited OGT covering oligonucleotide arrays and methods of using them to analyze polynucleotides The purchase of Axon Instruments Molecular Devices products does not convey any license under any of OGT s patent rights including any right to make or use oligonucleotide arrays under patents Customers may use Axon Instruments Molecular Devices pr
6. e Select a quicklist on the Quicklists tab e Select Apply Colors from the right mouse menu Data Windows When you open data from a microarray or a dataset it is displayed in a data window Once you have opened data into a data window you can open other windows displaying different views of the same data with the Window New Window command or you can open unrelated data in a new window with File Open Selected In New Window If you use Window New Window the windows are linked and so you can look at multiple views of the same dataset For example you might want to look at a Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 55 Self Organizing Map and a Principal Components Analysis of the same dataset and see how the clusters are plotted in the space of the principal components Because the windows are linked selecting a cluster also selects the same genes in the principal components analysis Data windows consist of two main panes Table pane top which contains five tabs and Views pane bottom which contains eight tabs Table Pane Although the top pane has five tabs Data Annotations Web Links Statistics and Advanced you do have the option of viewing tabs side by side To do this select View Split Substance Table to split the top view into two To continuing splitting the view into more panes keep selecting Split Substance Table After splitting into four pane
7. lowess normalization can be used in one of two ways to diagnose problems with experimental design and execution or to correct those problems in software If you are considering using lowess normalization you need to ask yourself e understand the physical basis of the defects that I am correcting e Could I perform this experiment with its systematic errors corrected and obtain the same results as I get from the lowess normalization of an experiment that has not had its systematic errors corrected If the answer to either of these questions is No then it would be wiser to perfect your experimental technique to remove intensity specific artifacts than to modify your data without clearly understanding the reasons for the modification If you do not understand the physical basis of what you are correcting then you can have no more confidence in the corrected data than in the uncorrected data Having chosen a normalization method it must be implemented in software Linear Ratio Normalization Because ratios are not normally distributed Acuity first takes the log of each ratio value when normalizing The mean x ofa set ofn ratios x is therefore Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 35 Ratio values less than 0 1 or greater than 10 can be excluded from the calculation as in GenePix Pro as can features flagged Bad Absent or Not Found To do this se
8. To swap branches e Select a large branch with the mouse so that one can see the obvious effects of the swap e From the right mouse menu select Swap Branches We are able to swap branches because swapping does not change the similarity of substances clustered under a node Similarity is plotted along the bottom of the Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 79 dendrogram and it is also reported in tooltips when you place the mouse over any part of the dendrogram We are interested in substances that are very similar so we will want to zoom in to some small portion of the dendrogram To zoom a branch e Select a branch with a high degree of similarity by clicking it with the mouse e Select Zoom Branches from the right mouse menu or use the Shift Z Hot Key Selecting the branch selected all the substances in the branch These are also highlighted in the Table pane and in all View tabs in the bottom pane of the window To view all selected substances in the dendrogram together in the Table pane e Ensure that you have a branch selected on the dendrogram e Select Data Group Selection or use the Hot Key To see a graph of all selected substances e Ensure that you have a branch selected on the dendrogram e Switch to the Graph tab and selected substances are graphed automatically No one hierarchical clustering method can tell you all there is to kno
9. a Web page devoted exclusively to comments and suggestions on how to improve Acuity If your computer is networked select the Help on the Web Send Feedback command to open a page on the Axon Instruments web site from where you can send a message to Axon or ask a question about Acuity Chapter 4 Technical Assistance e 89 Technical Assistance If you need help to resolve a problem there are several ways to contact Axon Instruments Molecular Devices World Wide Web Wwww axon com Phone 1 800 635 5577 Fax 1 510 675 6300 E mail axontech axon com Questions See Axon s Knowledge Base http support axon com Technical Assistance Customer License Agreement e 91 Customer License Agreement Customer License Agreement for Single User of Acuity 4 0 This software is licensed by Axon Instruments Molecular Devices Corp MOLECULAR DEVICES to you for use on the terms set forth below By opening the sealed software package and or by using the software you agree to be bound by the terms of this agreement MOLECULAR DEVICES hereby agrees to grant you a non exclusive license to use the enclosed MOLECULAR DEVICES software the SOFTWARE subject to the terms and restrictions set forth in this License Agreement Copyright The SOFTWARE and its documentation are owned by MOLECULAR DEVICES and are protected by United States copyright laws and international treaty provisions This SOFTWARE ma
10. and mixed in with a number of cell lines This blurring of the distinctions between samples can be lost in a simple cluster analysis but preserved in a scaling technique like Principal components Algorithm Details Principal Components Analysis is based on a standard decomposition in numerical analysis and is described in books on multivariate analysis such as Mardia et al 1979 Chapter 3 32 Analysis Algorithms Normalization Origins of variability Comparing the data from different array experiments is a non trivial task Small variations in the many steps that produce an array image can make comparisons across arrays problematic Variations can be due to differences in labeling efficiencies dye and batch variations chemical properties of different dyes pin tips slide batches and scanner settings e g red green channel settings multiple scanners Any of these variations can be corrected by normalization No one normalization method will correct all types of variation Choose a normalization method based on the known or expected sources of error and the characteristics of the experiment Validate the method empirically for example by reversing the dye labeling to test a normalization method for channel balancing In the acquisition step of an experiment one of the main contributors to variability between arrays is setting PMT values incorrectly so that the total signal acquired in one channel is significantly differe
11. another Different cluster techniques partition the data differently and so all partitions are arbitrary The disadvantage of PCA is that it rarely partitions the data into distinct sets there are few sets of substances in the PCA space that are completely separated from all other sets But that is the point of PCA it shows that clustering methods usually make arbitrary decisions about membership Which Clustering Method Should Use on My Data Users unfamiliar with clustering techniques usually want answers to the following questions e Which clustering method should I use e How many clusters should I find e Which similarity metric should I use Each of these questions presupposes the existence of a single best clustering method that will reveal all and only the intrinsic structure in the data There is no such method Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 15 The way to use Acuity is to use all the clustering and data reduction methods together By doing this you yourself will quickly discover the patterns in the data However some metrics are more appropriate for some analyses than others For K Means Analysis the Euclidean Squared metric usually the most appropriate as for that metric the cluster centroids are arithmetic averages of the points in each cluster Minimizing the Euclidean Squared distance of the cluster s points to the
12. between Principal Components Analysis and clustering is that in PCA each gene obtains a definite score on each component so that genes are ordered with respect to each component We can now select all three quicklists that we have created from the fold change condition t Test and PCA and create the intersection quicklist from these Chapter 4 78 Tutorial Hierarchical Clustering Hierarchical clustering along other clustering methods is a powerful way of looking at the global structure of a dataset To apply hierarchical clustering to a dataset e Ensure that you have a dataset open in the active data window e Select Clustering Hierarchical Clustering e The Hierarchical Clustering dialog offers a number of similarity metrics and linkage methods Accept the defaults e Click OK The Cluster Progress dialog box is displayed reporting the progress of the task When the progress reaches 100 the clustering result is added to the Project Tree under the dataset that was clustered Double click on the result in the Project Tree to display the result in the Visualizations tab Using Dendrograms The output of the hierarchical clustering algorithm is displayed in a visualization called a dendrogram It is important to understand that the structure as displayed in the Visualizations tab is not a unique representation of the mathematical clustering operation For this reason we are able to swap branches in the dendrogram
13. cluster s centroid naturally gives the centroid as the arithmetic average On the other hand for K Medians Analysis the cluster centroids are the medians of the points in each cluster Minimizing the City Block distance of the cluster s points to the cluster s centroid naturally gives the centroids as the median in this case Using Cluster Analysis Acuity includes both hierarchical and non hierarchical clustering methods Use these methods for different experimental tasks Use hierarchical clustering when you are interested in the relationship of each substance or array to every other substance or array For example if your experiment is an attempt to classify tumor subtypes from a large number of tissue samples where there is one tissue sample per array then you would use hierarchical clustering because you want to identify the tumor subgroups but you also want to see the degrees of similarity among the subgroups For example some tumor types might be subtypes of a more general tumor type You also use hierarchical clustering where you have little or no prior knowledge of how the data will be clustered as hierarchical clustering does not set the number of clusters to form before the analysis begins A disadvantage of hierarchical clustering is that it clusters substances into a single structure and pairs each substance with one other when several regulatory pathways may be present in biological systems and expressed substances
14. clusters that you have created e Quicklists tab lists all Quicklists lists of substances that you have created The Project Tree behaves like any other Windows tree You can cut copy paste drag and drop rename and delete items in the familiar way To create a new folder in the tree e Right click on the folder in which you want to create the new folder e Select New Folder from the right mouse menu To view and edit an item s properties e Select the item in the tree for example a microarray e Choose Properties from the right mouse menu Chapter 4 52 Tutorial e The microarray s properties are displayed in the Properties dialog box and are editable Microarrays Tab Microarrays are represented by one of two different icons to distinguish arrays with JPEGs from arrays without JPEGs e With JPEGs amp e Without JPEGs They can also be purple or orange which denotes their normalization status e Unnormalized B e Ratio normalized 4 In addition to these icons the Microarrays tab reports other information in columns next to the microarray names e The number of datasets in which the microarray is used e The number of features in the microarray e The normalizations that have been performed on the microarrays e The microarray parameters associated with the microarrays Datasets Tab A dataset is a set of spots Typically it consists of all the reliable spots from all the microarr
15. complete description of the Gene Shaving algorithm Principal Components Analysis Principal Components analysis is used to produce a low dimensional summary of the data It generates derived variables the components that are linear combinations of the results for different microarrays and which have maximal variance over substances subject to an orthonormality constraint Instead of looking at the expression profiles of 30 000 substances we examine the response patterns on a handful of components Each component is a linear combination of the substances We can examine two main quantities from Principal Components analysis the component loadings which show the values of the derived components for each microarray and the component scores which show the coefficients of each substance on the components Interpretation usually involves examination of both loadings and scores Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 31 Figure 1 Principal Components Analysis of 11 cancer cells lines on 68 microarrays from the Acuity PCA view The main advantage of Principal Components analysis over clustering methods is that it does not force us into a premature categorization of the data Figure 1 shows a plot of microarrays derived from many different cancer cell lines Some of the cancers such as leukemia and melanoma form into distinct clumps while others are spread out
16. directly in the Report tab or in an external web browser window To define a new web link e Open the Configure Web Links dialog box e Click the New button e Inthe Display Name field enter a name for the new web link this can be anything e Inthe URL field enter the URL for the web link If you include a substance name or ID in square brackets anywhere in the URL e g at the beginning or the end of the URL the name or ID is submitted directly to the web database Many databases have a specific syntax for such automated queries this is usually documented on the web site itself For the SGD database for example the URL with ID field 15 http genome www4 stanford edu cgi bin SGD locus pl locus ID You are not restricted to submitting the ID column you can put any substance property column name in the square brackets and submit it to the web based database For example if you have a column titled GI or EC numbers you can use GI or EC in place of ID assuming the web based database accepts those numbers Chapter 4 58 Tutorial Show hide and re order web links with substance properties in the Configure Columns Web Links dialog box Statistics Tab The Statistics tab displays basic statistics on columns in the current dataset For example if you have technical replicates i e replicate arrays and you re interested in the standard deviation or coefficient of variation across th
17. folder in the tree and then click OK to import the files Normalization The theory behind normalization is described in Chapter 3 so at this point we will discuss only the practical issues After importing your data into Acuity the first thing you need to do is normalize it If you have already created datasets on unnormalized microarrays you need to delete the datasets before performing normalization If you are working on the GPR files in the Microarrays tab of the demo database you need to delete all the demo datasets from the Datasets tab To normalize a set of microarrays Select the microarrays on the Microarrays tab of the Project Tree e Select Normalization Wizard from the right mouse menu e Select Ratio based normalization the default By default outlier features are not used to calculate normalization factors but all features are normalized e Click the Next button Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 65 e On the Summary page the normalization factors are displayed These should be close to 1 0 If they are far from 1 0 for example greater than 1 3 or less than 0 7 then you should consider scanning your microarray again as the PMT gain settings were not set very well e Click Finish e The normalization is performed and the Normalization Viewer is opened To see the effects of the normalization you can switch to Histogram mode and see how th
18. on another machine You do not need to have SQL Server 2000 installed either to install or to run Acuity in this mode For step by step instructions on the installation of Acuity please refer to the accompanying installation documents Connecting the Security Key The hardware protection key dongle that is shipped with Acuity can be attached to any computer on your network but we recommend that you attach it to the server computer For a local single user installation it is sufficient to attach the dongle to a USB port on the computer with Acuity For a multi user installation on a separate server computer you need to install network software to support the key This is explained in the Acuity installation document Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Installation 5 Starting Acuity After the successful installation of the software you will find the entry Axon Laboratory in your list of Programs in the Start menu and there will be two new icons on your desktop There is an Acuity 4 0 entry in your Axon Laboratory group and an icon on your desktop Both of these shortcuts will start Acuity Chapter 1 Introduction 7 Chapter 2 Introduction Acuity from Axon Instruments Molecular Devices is a fully featured microarray expression informatics software package that has the following features Analysis Hierarchical clustering with many differe
19. or higher e USB port Recommended Server Requirements Please discuss your server requirements with your computer vendor Server specifications depend strongly on the number of simultaneous users to support We recommend the following e Windows 2000 or 2003 Server operating system dual boot systems are not recommended e 768 MB RAM or more e SCSI or Firewire hard disk with 60 GB or more free for data storage e Back up device e g hard disk tape e USB port e Fast network card We do not recommend using Windows NT because it does not support the use of USB dongles For managing multiple users Windows 2003 Server has better security and stability than Windows 2000 Recommended Client Requirements The following is a recommended configuration for a client computer e IBM AT compatible computer with a 2 0 GHz or faster processor e Windows 2000 or Windows XP operating system dual boot systems are not recommended Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Installation 3 e 768 MB RAM or more e 1280x1024 display system with 16M colors e Internet Explorer 5 0 or higher e Fast network card As with all performance measures choose your system configurations based on the types of analyses that you perform For example hierarchical clustering uses large amounts of RAM on the client side while gene shaving uses large amounts of processor time If you are routinely open
20. that clusters are arranged on a two dimensional grid generated by two unobserved latent variables Clusters are constrained to have a regular arrangement on that grid The benefit provided by this is a simple representation of the similarity between clusters Clusters showing similar profiles across substances occupy nearby slots on the grid Clusters that are very dissimilar will occupy distant slots on the grid This goes some way to dealing with the inherent slop in cluster solutions and represents the potential uncertainty in classifying a substance into one cluster or a similar cluster Algorithm Details The algorithm used for Self Organizing Map analysis is based on the original algorithm developed by Kohonen 1990 The Self Organizing Map algorithm is very similar to the basic K Means algorithm initialize n cluster centroids assign elements to the closest cluster update the cluster centroids and iterate until the centroids are in stable locations Self Organizing Maps is a variant on Chapter 3 28 Analysis Algorithms the K Means algorithm in that the centroids are adjusted according to a weighting scheme after the update step The centroid adjustment step is included at every iteration For Self Organizing Maps the n centroids are initialized by randomly selecting n elements from the data set If the clustering is on substances then n substances are randomly chosen Because the clusters are initialized randomly if a S
21. that have signal in at least one channel we need to select both of these and then select the Apply OR button Each of these conditions needs to be successively constructed using the lists at the top of this dialog after which you click the Add To List button When you have all four conditions in the Combine Conditions pane select them and click Add To Query after which your query should look like this 532 Sat lt 3 AND F635 Sat lt 3 AND Flags gt 0 AND Ren R 635 532 gt 0 6 AND SNR 635 gt 3 OR SNR 532 gt 3 Click Next At this step of the Query Wizard we could filter further based on substance annotation for example we could select only the stress response genes However we are doing a global analysis so we can leave this page blank and click Next The Evaluate page of the Query Wizard reports the percentage of features that have matched our query You will have more or fewer spots depending on the quality of the arrays and the thresholds that you chose in your query If the percentage of features is acceptable click Finish Click OK in the dialog where you are asked to create or append You need to give the dataset a name and select a folder in the tree in which to save it and then click OK The dataset is opened in a new window Preparing a Dataset for Analysis The Query Wizard performs only spot specific filtering Once we have a dataset we may
22. the data that account for the most variance To perform a principal components analysis e Ensure that you have a dataset open in the active data window e Select Clustering Principal Components Analysis e Accept the defaults and click OK The Cluster Progress dialog box is displayed reporting the progress of the task When the progress reaches 100 the clustering result is added to the Project Tree under the dataset that was clustered Double click on the result in the Project Tree to display the result in the Visualizations tab This is a three dimensional scatter plot You can rotate the axes by clicking the mouse holding down the button and dragging Each of the axes can be thought of as representing an expression profile that explains variance in the dataset where the first component explains the most variance If you select Properties from the right mouse menu the components the amount of variance they explain and their profiles are displayed Looking at the scatter plot genes are plotted according to their similarity to the loadings profiles of the various components Chapter 4 76 Tutorial To select genes that score highly on the first principal component e Hold down the lt Alt gt key and drag a region around the points on the far right of the PCA scatter plot They are selected in red e Switch to the Profiles tab and you can see their profiles e If you do the same for points at the extreme left of the X ax
23. together with a clean database Please refer to Chapter 1 if you have not yet installed Acuity Starting Acuity When Acuity is installed a shortcut is copied to your Windows desktop To start Acuity double click this icon Connecting To A Database On starting Acuity the Welcome To Acuity login dialog box is displayed Use this dialog box to connect to a database before Acuity is opened To connect to a database e Configured databases are listed in the Data Source list e Select a database from the list enter your user ID and password and click OK On clicking OK Acuity is opened connected to your chosen database Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 49 Changing Your Password To maintain the security of your data in Acuity you should change your password periodically In particular if Acuity is installed with a blank password you should change it immediately To change your password e Login to Acuity e Open the Database Users dialog box e Select the user whose password needs to be changed and click the Properties button e Select the Change Password button and enter a new password Forgotten Passwords It is not uncommon for a user to forget the system administrator sa password If this happens you can change your system administrator password from the Welcome to Acuity login dialog box so long as you are an administrator on the comp
24. want to remove whole rows from it if they do not conform to further quality control criteria We may also want to transform the data in other ways to prepare it for analysis The commands to perform these operations are organized in the Analysis menu Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 69 Normalize to Column The first transformation to consider is called Normalize to Column in Acuity A common experimental design hybridizes the experimental sample with a pooled reference sample and so the ratios that are measured on the microarray are ratios relative to the pooled reference These are not biologically interesting The biologically interesting ratios are ratios of sample to sample In a time course experiment where we have used a pooled reference r the measured ratios on five microarrays are t r t r However the biologically interesting ratios might be something like 6 We transform the ratios from the first set to the second set using Normalize to Column which basically performs a division or subtraction for log ratio data e Select Analysis Normalize to Column e Select the microarrays to normalize typically this will be all the microarrays in the dataset e Select the microarray to normalize to In a time course experiment typically this will be the time zero microarray e Click OK Notice that the dataset in your Dataset t
25. 68 1998 Spellman P T Sherlock G Zhang M Q Iyer V R Anders K Eisen M B Brown P O Botstein D Futcher B Comprehensive identification of cell cycle regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization Molecular Biology of the Cell 9 3273 3297 1998 Hierarchical Clustering Rows and Columns Alizadeh A A Eisen M B Davis R E Ma C Lossos I S Rosenwald A Boldrick J C Sabet H Tran T Yu X Powell J I Yang L Marti G E Moore T Hudson J Jr Lu L Lewis D B Tibshirani R Sherlock G Chan W C Greiner T C Weisenburger D D Armitage J O Warnke R Staudt L M etal Distinct types of diffuse large B cell lymphoma identified by gene expression profiling Nature 403 503 511 2000 Perou C M Sorlie T Eisen M B van de Rijn M Jeffrey S S Rees C A Pollack J R Ross D T Johnsen H Akslen L A Fluge O Pergamenschikov A Williams C Zhu S X Lonning P E Borresen Dale A L Brown P O Botstein D Molecular portraits of human breast tumours Nature 406 747 752 2000 Chapter 3 44 Analysis Algorithms Ross D T Scherf U Eisen M B Perou C M Rees C Spellman P Iyer V Jeffrey S S van de Rijn M Waltham M Pergamenschikov A Lee J C Lashkari D Shalon D Myers T G Weinstein J N Botstein D Brown P O Systematic variation in gene expression patterns in human c
26. Acuity 4 0 MICROARRAY INFORMATICS SOFTWARE User s Guide Part Number 2500 0144 Rev F April 2005 Printed USA Copyright 2005 Axon Instruments Molecular Devices Corp No part of this manual may be reproduced stored in a retrieval system or transmitted in any form or by any means electronic mechanical photocopying microfilming recording or otherwise without written permission from Molecular Devices Corp QUESTIONS See Axon s Knowledge Base http support axon com VERIFICATION THIS PROGRAM IS EXTENSIVELY TESTED BEFORE DISTRIBUITON NEVERTHELESS RESEARCHERS SHOULD INDEPENDENTLY VERIFY ITS PERFORMANCE USING KNOWN DATA Verification Table of Contents e iii Table of Contents Chapter L Installation 1 Comiputer Requirements 1 Minimum Client or Server Requirement 1 Recommended Server Requirement 2 Recommended Client 2 ten Eek beret aes 3 Client and Database sso 4 4 Connecting the Security 4 Starting 5 Chapter 2 TntroduCtiOniscicisiscccccssccssesdsonscesssessousdsoncsessesseonsdsonssvsssessontssoonsessusssonceses 7 7 Ma Sali zat Ons a r n A E 9 Databasens E EE
27. ROARRAYS SEPARATELY LICENSED BY AFFYMETRIX NO LICENSE IS CONVEYED BY IMPLICATION ESTOPPEL OR OTHERWISE TO USE THIS INSTRUMENT WITH MICROARRAYS MADE USING IN SITU OR PHOTOLITHOGRAPHIC SYNTHESIS NO OTHER LICENSE IS CONVEYED BY IMPLICATION ESTOPPEL OR OTHERWISE UNDER ANY AFFYMETRIX PATENT OR OTHER INTELLECTUAL PROPERTY RIGHT This instrument is licensed by Affymetrix under the following patents U S Patent Nos 5 578 832 5 631 734 5 834 758 5 936 324 5 981 956 6 025 601 6 141 096 6 171 793 6 185 030 6 201 639 6 207 960 6 218 803 6 225 625 6 252 236 6 262 838 6 335 824 6 403 320 6 403 957 6 407 858 6 472 671 6 490 533 6 545 264 6 597 000 6 643 015 and 6 650 411 Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp
28. ainst a chosen microarray parameter For example the Profiles tab is where you display time course profiles of genes or gene expression profiles across all samples from an experiment Because selections in any Acuity view are linked to all views you can choose substances in the Data tab for example and then switch to the Profiles pane to see them graphed To select a microarray parameter apart from the current data type to graph on the X axis e Select Properties from the right mouse menu on the Profiles tab e Choose a new parameter from the list in the X axis group To zoom into any rectangular region of the graph e Select Zoom Mode from the View menu or the right mouse menu e Drag the region on the graph to be zoomed Plot Tab The Plot tab graphs any two columns of data from the top pane of a Data window So for example you can e Plot data from any two arrays against each other in order to see expression changes between them e Plot data from an array against an Annotation column such as gene length or chromosome position Chapter 4 60 Tutorial e Plot data from an array against a Statistics or Advanced column such as p value volcano plot To plot data e Switch to the Plot tab e Select the data to plot on the X axis from the X list e Select the data to plot on the Y axis from the Y list e Color the data by selecting a data type from the Color by list You can also draw two histograms on the Plot ta
29. ameters tab e Select the microarray parameters to display and click OK Summary Tab The Summary tab reports a summary of the current data source and the substance selected in the data window This can be useful for finding datasets and folders associated with microarrays Report Tab You can script your own analyses of the current data source in the Report tab Consult the Scripting Tutorial in the on line Help for more information on Acuity scripting To open an example Acuity Report click one of the hyperlinks on the default Report page or use the File Open Report command Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 63 The Report tab is also where Acuity web links are displayed Chromosome Tab The Chromosome tab draws genes on chromosomes for your chosen genome and can also plot gene expression levels directly on the chromosomes In order to construct a map of a genome you need to import a CDT file a tab delimited text file containing the chromosome coordinates for each gene in your genome of interest See the Acuity online Help for more details on the CDT file format Only a small number of major genomes have been well enough annotated to support the Chromosome tab Most of these are available from the UCSC genome browser download page http genome ucsc edu cgi bin hgText UCSC supports the following genomes Human chimp mouse rat chicken Fugu Dro
30. amines the behavior of substances on a small number of these components instead of the behavior across many microarrays Cluster Analysis is a grouping technique It reduces the complexity of datasets by partitioning data into a small number of sets The investigator can then examine the behavior of each set which is representative of the data in it instead of the behavior of each substance or microarray What Clustering Shows Consider clustering a dataset of 6000 substances into say 16 clusters You begin with 6000 different expression profiles and you end up with 16 representative expression profiles Yet within each cluster there are quite marked differences among expression profiles The clustering method ignores the differences It effectively throws away 5984 expression profiles and keeps 16 One hopes that the information thrown away is less useful that the information retained and highlighted If you ask the clustering algorithm to find 17 clusters instead of 16 then suddenly some substances that were in the same cluster are in different clusters Every clustering procedure tries to provide a summary of the dataset but it does this by throwing away information What is an argument for the efficacy of clustering If an organism has 6000 genes and one does an experiment on the organism the 6000 genes do not act independently On the contrary significant numbers of them are acting in concert Acuity 4 0 User s Guide Copy
31. ancer cell lines Nature Genetics 24 227 235 2000 K Means and K Medians Aronow B J Toyokawa T Canning A Haghighi K Delling U Kranias E Molkentin J D Dorn G W 2nd Divergent transcriptional responses to independent genetic causes of cardiac hypertrophy Physiological Genomics 6 19 28 2001 Brar A K Handwerger S Kessler C A Aronow B J Gene induction and categorical reprogramming during in vitro human endometrial fibroblast decidualization Physiological Genomics 7 135 148 2001 Soukas A Cohen P Socci N D Friedman J M Leptin specific patterns of gene expression in white adipose tissue Genes amp Development 14 963 980 2000 Tavazoie S Hughes J D Campbell M J Cho R J Church G M Systematic determination of genetic network architecture Nature Genetics 22 3 281 5 1999 Self Organizing Maps Huang Q Liu doN Majewski P Schulte leAC Korn J M Young R A Lander E S Hacohen N The plasticity of dendritic cell responses to pathogens and their components Science 294 870 875 2001 Saban M R Hellmich H Nguyen N B Winston J Hammond T G Saban R Time course of LPS induced gene expression in a mouse model of genitourinary inflammation Physiological Genomics 5 147 160 2001 Tamayo P Slonim D Mesirov J Zhu Q Kitareewan S Dmitrovsky E Lander E S Golub T R Interpreting patterns of gene expression with self organizin
32. and substance i is positive We also know thata b c d p From such a table one can generate a number of distance measures Simple Matching dy b e p Chapter 3 20 e Analysis Algorithms This is the ratio of mismatches to the total number of values Jaccard dj b c a btc This is the ratio of mismatches to the total number excluding joint absences Bray Curtis dj b c 2at b c This is the ratio of mismatches to the total number weighted to joint matches excluding joint absences Some Background Information on the Metrics Geometric Metrics Both the Euclidean Squared metric and the City Block metric have geometric interpretations Loosely speaking the Euclidean Squared Metric gives clusters that are sphere shaped while the City Block metric gives clusters that are diamond shaped For K Means Analysis the Euclidean Squared metric usually the most appropriate as for that metric the cluster centroids are arithmetic averages of the points in each cluster Minimizing the Euclidean Squared distance of the cluster s points to the cluster s centroid naturally gives the centroid as the arithmetic average On the other hand for K Medians Analysis the cluster centroids are the medians of the points in each cluster Minimizing the City Block distance of the cluster s points to the cluster s centroid naturally gives the centroids as the median in this case Population Metrics The Bray Curtis and the Canb
33. ap from the right mouse menu To show the average profile only select Graph from the right mouse menu To view the distribution of a Self Organizing Maps cluster on a Hierarchical Cluster display e Select a cluster on the Self Organizing Maps display e Double click on a hierarchical cluster result in the Project Tree to open it e The substances from the Self Organizing Maps cluster are highlighted in the hierarchical cluster Chapter 4 84 Tutorial To view a hierarchical cluster and Self Organizing Maps together e Open a hierarchical cluster analysis result e Select Window New Window to open a new window on the same data source e Double click on a Self Organizing Maps analysis result to open it into the new window e Select Window Tile Horizontal to tile the windows e Expand the displays by clicking on each analysis result in turn and selecting View Expand or use the Ctrl E Hot Key To find the cluster in which a substance is e Select the substance in the Table view e 5 cluster is selected in the Visualizations tab Visualizing Principal Components Analysis and SOMs Together You will often find that the first principal component corresponds closely to a cluster produced by the Self Organizing Maps algorithm To view Principal Components and Self Organizing Maps together e Open a Principal Components Analysis result e Select Window New Window to open a new window on the same data source
34. at the shape Apart from Gene Shaving non hierarchical clustering is very fast and uses little memory It is therefore suitable for large datasets say gt 100 microarrays Gene Shaving is unsuitable for very large datasets Hierarchical Cluster Analysis Hierarchical cluster analysis produces the familiar tree structures called dendrograms The hierarchical nature of the tree means that clusters that are not linked together at one degree of similarity are linked together at a lesser degree of similarity Eventually all objects in the tree are linked together Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms e 17 In a hierarchical cluster the number of clusters is not set before the analysis it is derived from the analysis To create a hierarchical cluster one must specify a similarity metric and a linkage method The similarity metric is used to form data points into clusters while the linkage method is used to join clusters to form a tree Similarity can be expressed mathematically in many different ways Acuity employs three different classes of similarity metrics based on correlation coefficients distance measures and binary measures Binary measures tend to produce trees with much less structure than those produced by either correlation coefficients or distance measures Large numbers of substances are grouped together at the same degree of similarity This has a numbe
35. austive groups that is every observation is in one and only one group The number of groups is chosen a priori The benefit of using K Means analysis is that instead of looking at the response of say 30 000 substances across arrays we look instead at the response of a much smaller set of clusters perhaps ten or so It may be informative to examine which substances have been allocated to which cluster For large numbers of substances K Means is very much faster than a hierarchical cluster analysis K Means works by optimizing a quality criterion generally involving the within cluster sums of squares Since the problem of allocating a large number of substances 30 000 to a small number of clusters 10 is a huge combinatorial task the algorithm works with a set of heuristics This means that we can be fairly confident of obtaining a reasonable solution but have little chance of obtaining the best solution Since however the best solution is dependent on the optimality criterion and there are no compelling reasons for choosing one criterion over another the best solution is not really a meaningful target The consequence of these considerations is that there is no right or wrong solution using K Means There are many different variants of the algorithm and they will be more or less useful under different types of pathological data Users must never be uncritical in their acceptance of K Means results but the procedure will freque
36. aximizes the difference between each dimension If one has columns in a purely random dataset each column explains 100 n of the variance of the dataset When looking at a PCA result therefore the only significant components are those that explain more than 100 n of the variance In the case of the seven Diauxic arrays this number is 100 7 or about 14 Double click on the PCA display to see the variance explained by each component What we look for in a PCA display are points clumped together By being together they have similar expression profiles by being slightly separated from other points they form a distinct group Chapter 3 14 Analysis Algorithms PCA or Clustering The difference between Principal Components Analysis and clustering is that there is much less arbitrariness in PCA While a clustering technique always has to make a somewhat arbitrary decision about which cluster to assign a substance to Principal Components Analysis is more likely to produce an informative representation of the real structure of the data Why is clustering arbitrary Fundamentally each clustering technique puts similar substances together so each clustering technique must decide on some mathematical measure of similarity One reason why there are so many different clustering techniques is that there are many different measures of similarity Furthermore there is no sense of any one measure of similarity being better than
37. ays in a single experiment To create and open a dataset containing all the spots from a set of microarrays e On the Microarrays tab of the Project Tree select the microarrays that you wish to analyze e From the right mouse menu select Create Dataset From Selection Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 53 Datasets can be as small or as large as you like from several spots to all the spots in the database Datasets are the units on which most major Acuity analyses are performed see below Performing Analyses Each time you do a cluster analysis for example the cluster result is listed in the Project Tree under the dataset from which it was created By default the data type i e the GPR column displayed for each microarray used in both microarrays and datasets is Log Ratio To change the data type of the current dataset e Select the Configure Current Data Type to Retrieve command e Select a data type to view and click OK Alternatively the current data type is displayed in a list box at the top of the data window You can select a new data type to retrieve from this list Creating a dataset from all the spots on a microarray as described above is not what we do when analyzing a real experiment To do an actual data analysis we use Acuity s Query Wizard to extract just the reliable spots from our set of experimental arrays This is described belo
38. b e Switch to the Plot tab e Select the data to histogram on the X axis from the X list and select the X Histogram button e Select the data to histogram on the Y axis from the X list and select the Y Histogram button Visualizations Tab The Visualizations tab displays analysis results such as e Dendrograms e Various non hierarchical clusters and e Principal components analyses See Performing Analyses below Features Tab The Features tab is a GPR viewer for the column selected in the Table view It functions very much like the GenePix Pro Image Results and Scatter Plot tabs Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 61 By default the Features tab displays only small number of GPR columns from the selected microarray To download more columns from the database e Select a microarray in the Table view of the top pane and switch to the Features tab e Use lt F5 gt to retrieve data from the selected column e Select Configure Columns from the right mouse menu on the Features tab table e Select the GPR columns to display in the Features tab and click OK To graph two data types against each other in the Features tab e Select all the substances to plot e Select the data to plot on the X axis from the X list e Select the data to plot on the Y axis from the Y list e Color the data by selecting a data type from the Color by list Becaus
39. b see above select Advanced Transform Advanced Columns e Select Student s t Test from the list of Columns to transform e Inthe X field select log x log 2 e Click OK Chapter 4 74 Tutorial We use log p because the p values are distributed over many orders of magnitude Student s t Test equal variances F7_Diauxic Figure 3 Volcano plot To plot log ratio against log p Click on the Plot tab From the X control at the top of the Plot tab select F7_Diauxic From the Y control at the top of the Plot tab select Student s t Test From the Color By control select F7_Diauxic Select all genes in the dataset by clicking in the Data pane at the top and using lt Ctrl gt lt A gt Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 75 The resulting scatter plot is called a volcano plot because the distribution often looks like a volcano erupting The interesting thing about the volcano plot is that one can select features that have both large fold change and are statistically significant look at the genes that fall in the top left or top right parts of the distribution By Principal Components Analysis In principal components analysis entirely new variables the components are derived from the data and substances are plotted in the space defined by these variables The components can be thought of as corresponding to the dimensions in
40. bstances Select Create Quicklist from the right mouse menu to create a quicklist Select Create Dataset From Selection from the right mouse menu to create a dataset of the spots Chapter 4 82 Tutorial You can also match the expression of a user defined profile To do this e Select Advanced Match Expression on User Profile e In this dialog you can draw a profile in which you are interested On clicking OK the main Data table is sorted according to the correlation to that profile If visualizing the results with a Color Map Only dendrogram select Auto Sort Color Map from the right mouse menu to sort the color map to the sorted order of the correlation coefficients Non hierarchical Clustering Non hierarchical clustering partitions substances into unrelated sets so that membership of one set does not necessarily imply membership of any other set K Means K Medians and Self Organizing Maps clusters are mutually exclusive while the Gene Shaving algorithm allows substances to belong to more than one cluster K Means K Medians and Self Organizing Maps use essentially the same algorithm except that K Means and K Medians produce an unordered list of clusters while Self Organizing Maps organizes the clusters on a 2 dimensional grid according to their relative similarity It is therefore always more informative to use Self Organizing Maps instead of K Means or K Medians To analyze a dataset using Self Organizing Maps e En
41. can participate in more than one pathway Chapter 3 16 Analysis Algorithms Hierarchical clustering is relatively slow method and it requires a very large amount of computer memory RAM compared to the non hierarchical techniques Use non hierarchical clustering when you are interested in separating substances into distinct classes but you are not as interested in relationships between the classes For example you might be interested in pathogen induced expression across a genome and you want to identify groups of genes by function In sucha case the functions of different groups may not be related so there is no sense in looking at the similarity of different groups As non hierarchical clustering forces the data into a user defined number of clusters you also use it when you have some a priori idea about the number of clusters that you want to form For example you might think that the response to the pathogen occurs in three main phases say early middle and late and you want to see genes clustered into these three temporal groups Alternatively you may perform a hierarchical cluster analysis to identify the number of main clusters in a sample and then instruct the non hierarchical clustering method to find that many clusters Gene Shaving is unique among clustering techniques because it groups together both positively and negatively correlated substances That is it ignores the sign of a correlation and looks only
42. d to examine your cluster solutions to see if any matches the profiles in Figure 5 You may need to perform more clusters with various algorithms before you find what you are looking for e gene names in Figure 5 that are associated with the expression profiles are listed in the Gene column of the substance properties pane Once you think you have found the right expression profiles look in the Gene column to see if the genes are in it Finding the expression profiles and identifying the genes is not a simple step by step procedure You may have to cluster your data using several algorithms and with several different sets of options for each algorithm before you find the profiles in Figure 5 This is exactly the way that microarray experiments are performed Summary This is the end of the tutorial We hope that you enjoy using Acuity Axon Instruments Molecular Devices has put every effort into designing and constructing an application that will work with you to get your microarray informatics tasks done efficiently Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 87 Remember to refer to the on line Help and the rest of the Manual if you have any further questions about using Acuity If you encounter a problem that you can t solve don t hesitate to contact Technical Support Feedback Axon Instruments Molecular Devices welcomes feedback on all its products There is
43. e histogram has been shifted to be centered on zero When your microarrays are ratio normalized all relevant columns in the microarray are modified in the database Therefore whenever you analyze the data from that microarray you are using the ratio normalized data Let us also perform a Lowess normalization e Select the microarrays on the Microarrays tab of the Project Tree Select Normalization Wizard from the right mouse menu e Select Lowess Slide Normalization e Click the Finish button e The normalization is performed and the Normalization Viewer is opened showing before and after M versus A plots Unlike ratio based normalization when a Lowess normalization is performed a new data type new column in the GPR file is created in the database containing the Lowess normalized log ratio By default this is called Lowess Log Ratio We create a new data type because Lowess normalization is not reversible and we do not want the original data in the database to be irreversibly affected To use the Lowess normalized data for your analysis you need to select Lowess M Log Ratio as your data type before beginning your analyses Chapter 4 66 Tutorial Creating a Dataset Datasets are the units of analysis in Acuity Typically a dataset consists of all the reliable data from a set of microarrays that together form an experiment There are two main reasons why we might create a dataset from only a subset of the avai
44. e rule a case is joined to an existing cluster if it has the same level of similarity as an arithmetic average of the levels of similarities of all members of the existing cluster In other words the distance between average neighbors determines the distance between clusters This method tends to be equally good with data that are in long chains or in clumps Chapter 3 22 e Analysis Algorithms Single Linkage Single linkage forms clusters according to the rule a case is joined to an existing cluster if it has the same level of similarity as at least one member of the existing cluster In other words the distance between nearest neighbors determines the distance between clusters This method tends to produce long chain shaped trees Complete Linkage Complete linkage forms clusters according to the rule a case is joined to an existing cluster if it is within a certain level of similarity to all members of the existing cluster In other words the distance between furthest neighbors determines the distance between clusters This method tends to produce trees where clusters are clumped together so it may not be appropriate if the data are in fact in long chains Clustering Substances and Arrays Acuity gives you the options of clustering substances or arrays or both substances and arrays in the one process You would never cluster both substances and arrays in a time series experiment because the arrays in such an e
45. e the Features tab lists individual spots as opposed to spot averages for individual substances as in the main Table view if you select a substance in the Table view all its replicates are selected in the Features tab The images displayed in the Features tab are the JPG images that are imported with GPR files When you select rows in the Features tab table they are automatically selected and zoomed on the image Parameters View The Parameters tab displays all defined microarray parameters and their values Like substance properties microarray parameters can be imported from a tab delimited text file Chapter 4 62 Tutorial To import microarray parameters e Select the File Import Other Microarray Parameters command e Select an MDT file MDT files are tab delimited text files that contain a row of column titles a column of microarray names and other columns of annotations There is a sample MDT file on the Acuity installer CD in the Sample Data Diauxic directory You may like to import this file as we will use it later in the sample experiment You can edit microarray parameters manually right click on the tab and select Properties to open the Microarray Properties dialog box where you can edit the parameters of the selected microarray To show or hide microarray parameter columns e Open the Configure Columns Microarray Parameters dialog box or select Configure Columns from the right mouse menu on the Par
46. elf Organizing Maps analysis is run on the same dataset twice the results will be slightly different As with K Means and K Medians this demonstrates the important fact that cluster membership is always somewhat arbitrary When repeating a cluster analysis one looks for the substances that do not shift from one cluster to another If the number of clusters is assumed to be non prime then each cluster can be mapped to a location in a rectangular two dimensional latent hidden variable space At each iteration of the algorithm a cluster centroid is adjusted to be a weighted average of the neighboring centroids in latent variable space The weights of the neighboring centroids are dependent on the number of elements in the respective cluster and the distance between the clusters in latent variable space A cooling schedule is employed so that the apparent distance between the centroids in latent variable space increases to infinity so that the eventual influence of neighboring clusters tends to zero as the algorithm iterates The clustering results can be viewed on a plot of the clustering elements and groups in the latent variable space This plot is important because the algorithm is constructed so that elements in neighboring clusters in latent variable space should be similar For example a data set containing two main classes of substances may be clustered on substances into 16 clusters using Self Organizing Maps Although the clust
47. er normal use For 90 days from the date of receipt MOLECULAR DEVICES will repair or replace without cost to you any defective products returned to the factory properly packaged with transportation charges prepaid MOLECULAR DEVICES will pay for the return of the product to you but if the return shipment is to a location outside the United States you will be responsible for paying all duties and taxes Before returning defective products to the factory you must contact MOLECULAR DEVICES to obtain a Service Request SR number and shipping instructions Failure to do so will cause long delays and additional expense to you MOLECULAR DEVICES has no control over your use of the SOFTWARE Therefore MOLECULAR DEVICES does not and cannot warrant the results or performance that may be obtained by its use The entire risk as to the results and performance of the SOFTWARE is assumed by you Should the SOFTWARE or its documentation prove defective you assume the entire cost of all necessary servicing repair or correction Neither MOLECULAR DEVICES nor anyone else who has been involved in the creation production or delivery of this SOFTWARE and its documentation shall be liable for any direct indirect consequential or incidental damages arising out of the use or inability to use such products even if MOLECULAR DEVICES has been advised of the possibility of such damages or claim This warranty is in lieu of all other warranties expressed or implied
48. ere we have single channel data For the case of two arrays the Lowess process robustly fits a smooth trend function to the data and then subtracts the trend from the from the original array data For more than two arrays the situation 15 slightly trickier each pair of arrays has an associated trend function Furthermore each array is paired with every other array The basis of the method is to repeatedly construct the trend curves for every pair of arrays and to subtract each trend curve from all affected array pairs Two or three cycles through every pair of arrays may be required for convergence Overall the Cyclic Lowess Normalization procedure robustly normalizes the array data while retaining significant quantitative information The cost of this procedure is that it is demanding computationally Lowess is a fairly expensive technique when applied to large data sets and the cyclic part ensures it is applied some multiple of times where is the number of arrays in the data set Quantile Normalization Quantile Normalization makes the probes on each array have the same distribution by translating the cumulative distribution function of probes to a standard distribution function The standard distribution is computed by ranking the probe values in each array and then computing the average value for each rank The original probe values are replaced by the average for the probe s rank For example consider the case of severa
49. ering results may suggest that the substances fall into many more than two groups it may be found that the clusters occur in two general regions in latent variable space For example there may be 5 clusters containing substances in the top left of the latent variable space and 3 clusters containing substances in the bottom right of the latent variable space This grouping pattern could only be discovered by plotting the clustering results in the latent variable space Note that such a Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 29 grouping is invariant to a transposition in the two dimensional latent variable space That is the regions would still appear distinct if the horizontal and vertical axes of the latent variable plot were reversed The Self Organizing Maps algorithm in Acuity 4 0 uses a similar set of metrics as are available for hierarchical clustering For more information on the metrics see their description above in the Hierarchical Clustering section Gene Shaving Gene Shaving falls into the category of non hierarchical cluster analysis along with K Means K Medians and Self Organizing Maps Gene Shaving is a novel cluster analysis technique developed by Hastie et al 2000 especially for expression analysis Its aim is to identify groups of substances genes that have coherent expression and are optimal for various properties of the variation in their expr
50. erra metrics were originally devised for measuring populations in different habitats As such these metrics are Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms e 21 most applicable when the data points have non negative values as would be the case for population counts For these metrics the p elements in a point are the population counts for different species Comparison of two points is in effect the comparing of the different population counts species by species Binary Metrics These metrics are applicable in the case of binary data where for instance the presence of an attribute is signaled by a 1 and the absence by a 0 The Simple Binary Matching metric simply records the number of times that the elements of one point differ from those of another It gives equal weighting to the case of an attribute being present in both points and the case of an attribute being absent in both points For cases where we are more interested in the common presence of an attribute than in its absence the Jaccard metric is the preferred metric This is reflected in the fact that it does not use the d value in its calculation Linkage Different linkage methods can produce clusters with very different shapes Choose a linkage method based on any a priori structure in the data or experiment with different linkage methods Average Linkage Average linkage forms clusters according to th
51. es Annotations Tab The Annotations tab displays substance annotations Substance annotations are imported to Acuity from plain text files To import substance annotations e Select the File Import Other Substance Annotations command e Select an SDT file SDT files are tab delimited text files that contain a row of column titles a column of substance IDs and other columns of annotations There is a sample SDT file on the Acuity installer CD in the Sample Data Diauxic directory You may like to import this file as we will use it later in the sample experiment For more detail on the SDT file format and where to obtain annotation data see the Import Substance Annotations topic in the Online Help To show or hide substance property columns use the Configure Columns Substance Annotations dialog box Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial e 57 If you have gene ontology properties such as Component Function and Process you can create gene ontology quicklists from the substance properties pane using Analysis Quicklist and Coloring Operations Create Quicklist From Substance Annotations Web Links Tab The Web Links tab displays URLs that link the genes in your dataset directly to online genomics databases You can submit the various IDs and gene names that you have in the Annotations tab to online databases and have the results of those queries displayed
52. ession The algorithm as implemented in Acuity is constrained to produce high variance clusters and high coherence between members of each cluster Gene Shaving differs from other cluster algorithms in a number of interesting ways The clusters of substances are constructed to show large variation across the set of arrays That is they are likely to contain a strong differential expression signal The clusters of substances are not exclusive A substance may be allocated to more than one cluster Cluster profiles are independent of each other The sign of a substance s contribution to a cluster is potentially arbitrary That is a substance showing linear increase along the arrays is likely to be clustered with a substance showing linear decrease along the arrays You can also use the Absolute Pearson Correlation in Hierarchical Clustering to get both correlated and anti correlated substances clustered together Chapter 3 30 e Analysis Algorithms Gene Shaving is relatively fast for a large number of substances but the cost increases rapidly with the number of arrays It is typically slower than K Means One constraint of Gene Shaving is that it cannot produce more clusters than there are arrays In the extreme situation of only two arrays at most two clusters are found Gene Shaving may on occasions reveal structure that is not apparent in more traditional cluster algorithms Algorithm Details See Hastie et al 2000 for a
53. g maps Methods and application to hematopoietic differentiation PNAS 96 2907 2912 1999 Principal Components Analysis Alter O Brown P O Botstein D Singular value decomposition for genome wide expression data processing and modeling PNAS 97 10101 6 2000 Hilsenbeck S G Friedrichs W E Schiff R O Connell P Hansen R K Osborne C K Fuqua S A Statistical analysis of array expression data as applied to the problem of tamoxifen resistance Journal of the National Cancer Institute 91 5 453 9 1999 Holter N S Mitra M Maritan A Cieplak M Banavar J R Fedoroff N V Fundamental patterns underlying gene expression profiles simplicity from complexity PNAS 97 15 8409 14 2000 Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 45 Raychaudhuri S Stuart J M Altman R B Principal components analysis to summarize microarray experiments application to sporulation time series Pacific Symposium on Biocomputing 455 466 2000 Tamayo P Slonim D Mesirov J Zhu Q Kitareewan S Dmitrovsky E Lander E S Golub T R Interpreting patterns of gene expression with self organizing maps Methods and application to hematopoietic differentiation PNAS 96 2907 2912 1999 Chapter 3 Tutorial 47 Chapter 4 Tutorial Introduction This tutorial is in two main sections The first half of the tutorial is an extended t
54. gorithms New York John Wiley amp Sons Inc 1975 Hartigan J A amp Wong M A A K means clustering algorithm Algorithm AS 136 Applied Statistics 28 126 130 1979 Hastie T Tibshirani R Eisen M B Alizadeh A Levy R Staudt L Chan W C Botstein D amp Brown P Gene shaving as a method for identifying distinct sets of genes with similar expression patterns Genome Biology 1 2 research0003 1 0003 21 2000 Hastie T Tibshirani R Friedman J The Elements of Statistical Learning Data Mining Inference and Prediction New York Springer 2001 Linde Y Buzo A amp Gray R M An algorithm for vector quantizer design IEEE Transactions on Communications 28 1 84 95 1980 Kohonen T The self organising map Proceedings of the IEEE 78 9 1464 1480 1990 Mardia K Kent J and Bibby J Multivariate Analysis Academic Press 1979 Tamayo P Slonim D Mesirov J Zhu Q Kitareewan S Dmitrovsky E Lander E S Golub T R Interpreting patterns of gene expression with self organizing maps Methods and application to hematopoietic differentiation PNAS 96 2907 2912 1999 Normalization Dudoit S Yang Y H Callow M J Speed T P Statistical methods for identifying differentially expressed genes in replicated cDNA microarray experiment Statistica Sinica 12 111 139 2002 Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analys
55. group comparisons by one way ANOVA Support for dye swap microarrays in datasets Column arithmetic on any data column Multiple column transformations on datasets row and column centering and scaling Image display and integration with data tables scatter plots and all other visualizations in Acuity Lasso selection on images User definable flag values Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Introduction 9 Scripting engine for customizable analysis through VBScript JavaScript or ActiveX objects Analysis queuing Fully integrated with GenePix Pro Web links for unlimited access to web based databases including pathways Create datasets from completely general database queries across all microarrays and all annotations in the database Construct ontologies from imported gene ontology information Merge microarrays Apply GAL file to microarrays Visualizations Dendrograms 2 D interactive plots Animated interactive 3 D scatter plots Chromosome visualization Normalization Viewer shows unnormalized and normalized data in the same window in scatter plots or histograms M vA plots including lowess print tip smoothing curves Line graphs of any microarray parameter Scatter plots of any GPR or other microarray data type or any analysis data type such as p value or correlation coefficient Color tables Chapter 2 10 Introduction Export any
56. ing large datasets say more than 200 microarrays but using fast analyses like self organizing maps or K Means a fast hard disk on the server and fast server access is more valuable then a fast processor on the client Installation For complete step by step installation instructions please consult the accompanying Acuity installation document Acuity consists of both client software and database software Before installing the Acuity database you need to install Microsoft SQL Server 2000 which you purchase independently of Acuity Once SQL Server 2000 is installed proceed with installing Acuity The Acuity installation CD also includes MSDE the free single user version of Microsoft SQL Server 2000 You can use MSDE in place of SQL Server 2000 for a single user installation MSDE has a database size limit of 2 GB To run the Acuity installer double click setup exe on the Acuity CD Alternatively from the Start menu select and type x setup exe where x is the drive letter of your CD ROM drive Chapter 1 4 Installation The install program offers the following installation options Client and Database Select this option if you want this machine to be a database server and to run Acuity You need to have SQL Server 2000 already installed and running Client Only Select this option if you want to install Acuity on this machine but not the database You will have to connect to a database
57. is you see that they have the same profile but reflected in the X axis To select these differentially expressed genes a little more rigorously e Right click on the principal components analysis display and select Component Scores e Click the Add button to save them to the Advanced tab e On the Advanced tab you can sort them and choose for example the top 20 and the bottom 20 genes in the list and create a dataset quicklist from them As we did with the volcano plot above you can plot principal component score on the X axis versus p value on the Y axis and obtain an even more informative volcano plot Instead of looking at the log ratio on one microarray one is looking at the first principal component which represents the most variation in the dataset Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 77 Student s t Test equal variances PCA 6 components Component 1 Figure 4 Volcano plot The advantage of Principal Components Analysis over a simple fold change filter or even a t Test is that genes are organized both by variance and by profile e Genes far from the origin are changing more e Genes closer to the different axes are changing with different profiles So whereas a fold change filter and a t Test would group together genes with different profiles Principal Components Analysis separates them While this may sound like clustering the difference
58. is Algorithms 43 Yang Y H Dudoit S Luu P Lin D M Peng V Ngai J Speed T P Normalization for cDNA microarray data a robust composite method addressing single and multiple slide systematic variation Nucleic Acids Research 30 4 e15 2002 Robust Multichip Analysis RMA B M Bolstad R A Irizarry M Astrand and T P Speed A Comparison of Normalization Methods for High Density Oligonucleotide Array Data Based on Variance and Bias Bioinformatics 19 185 193 2003 Irizarry RA Hobbs B Collin F Beazer Barclay YD Antonellis KJ Scherf U Speed TP Exploration Normalization and Summaries of High Density Oligonucleotide Array Probe Level Data Biostatistics Vol 4 Number 2 249 264 2002 Rafael A Irizarry Benjamin M Bolstad Francois Collin Leslie M Cope Bridget Hobbs and Terence P Speed Summaries of Affymetrix GeneChip probe level data Nucleic Acids Research 31 4 e15 2003 Use of Algorithms for Gene Expression Analysis Hierarchical Clustering Rows Only Iyer V R Eisen M B Ross D T Schuler G Moore T Lee J C Trent J M Staudt L M Hudson J Jr Boguski M S Lashkari D Shalon D Botstein D Brown P O The transcriptional program in the response of human fibroblasts to serum Science 283 83 87 1999 Eisen M B Spellman P T Brown P O Botstein D Cluster analysis and display of genome wide expression patterns PNAS 95 14863 148
59. is is to click the Select From Folder button and select the microarrays there Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 67 e Weare constructing query across a number of steps in this wizard so we need to click the Add To Query button to add this first criterion to our query e Click Next to get to the next step of the wizard e The second step of the Query Wizard is where we apply quality control conditions on our spots All GenePix Pro data types are listed in the Parameter column Let s apply the following filters to our dataset We include only the following spots Spots with only a small percentage of saturated pixels Spots that are not flagged bad nor found or absent Spots with relatively uniform intensity and uniform background Spots that are detectable above background The first criterion requires us to construct two conditions F635 Sat lt 3 F532 Sat lt 3 The second criterion can be applied with this condition Flags gt 0 The third criterion can be applied with this condition Rgn R2 635 532 gt 0 6 0 6 15 a recommended threshold but you could use 0 5 or 0 4 if too many features are failing or 0 7 or higher if you want the filter to be more stringent The last criterion can be applied with the two signal to noise ratio data types SNR 635 gt 3 SNR 532 gt 3 Chapter 4 68 Tutorial However because we want spots
60. l arrays and consider further the minimum probe value on each array The average of these minimum probe values is the estimate of the minimum value of the standard distribution The minimum value in each array is then replaced by this average The same procedure is done for the second smallest value in each array and so on until all original values are replaced by their ranked equivalent value Chapter 3 40 Analysis Algorithms Quantile Normalization achieves statistical robustness by the use of the ranking process and retains some of the original quantitative information via the averaging process Since ranking is a relatively fast procedure namely O Nlog N in the number of probes N the overall computational cost of Quantile Normalization is considerably lower than Cyclic Lowess Normalization Summarization The two Summarization algorithms provided in RMA in Acuity are Median Polish and Robust Linear Model RLM Both algorithms estimate the expression value for each probe set by solving a statistical linear model for the probe set data For a single probe set the model takes the form Yig wt In this model i indexes the probe value within the given probe set and j indexes the Affymetrix chip The y j represents the logarithm of the probe value the represents the probe effect and the represents the chip effect The sij represents statistical noise within the data and finally u is the overall mean value
61. lable data instead of from each feature from every microarray in an experiment e We remove unreliable data from the dataset For example we remove data points derived from slide defects such as smears e We remove uninteresting data from the dataset For example we may have control features used for normalization that are not needed for downstream analysis or we remove substances that do not show any interesting behavior in order to make the analysis task more tractable Let us concentrate on removing the unreliable data This can be a treacherous task due to the subjective nature of what counts as good data the variability in data quality across microarrays the lack of accepted standards for good data and the problem of translating image based defects into numerical conditions on array data types The easiest way to do this and it is relatively easy is to make a list of common feature and slide defects and then translate them into numerical conditions on GenePix Pro and Acuity data types Note that all these conditions should be applied to microarrays that have already been normalized We apply the quality control conditions and create a dataset in the Acuity Query Wizard e Select Analysis Create Dataset From Query Wizard e At the first step of the Query Wizard we need to select the microarrays from which we are going to create a dataset If all the microarrays are in the same folder then the easiest way to do th
62. lect the GenePix Pro Settings options in the Normalization Wizard Four color normalization is done on a ratio by ratio basis For example if you choose to normalize so that the mean of the Ratio of Medians is set to 1 0 each Ratio of Medians data type that you have defined on the microarray is normalized independently To calculate the wavelength specific normalization factors that are reported in the Normalization Viewer the change to the ratio is distributed equally between the wavelengths so one wavelength scales up by the square root of the ratio scale factor and the other scales down For example suppose the mean of Ratio of Medians of 635 532 is 1 21 The square root of 1 21 is 1 1 so we set the scale factor for the 532 wavelength to 1 1 and the scale factor for the 635 wavelength to 1 1 1 0 91 After applying these scale factors the new mean of the Ratio of Medians is 1 21 0 91 1 1 1 0 Because normalization can scale up the data it is possible for normalization to produce pixels with intensities greater than the hardware limit of 65535 Lowess Normalization As described above Lowess normalization is non linear so features with different intensities are normalized differently The easiest way to see this effect is to look at a Lowess normalization in the Normalization Viewer with lowess curves displayed such as in Figure 2 The top pane shows the unnormalized data and the bottom pane shows the normalized data B
63. make one fundamental assumption Genes that are expressed together share common functions From this assumption we infer the following which is sometimes called guilt by association We can suggest possible roles for genes of unknown function based on their temporal association with genes of known function For experiments in which the microarrays are derived from different tissue samples instead of the same sample at different times we can formulate the guilt by association statement as We can categorize samples of unknown physiological state based on their association with samples of known physiological state The central analytical task then is to group together substances or microarrays based on the similarity of their expression profiles This is we want to reduce the Chapter 3 12 Analysis Algorithms complexity of the data so that large scale trends and structure are revealed Once we have a sense of the large scale structure we can investigate the fine structure in these trends further There is a wide variety of well known mathematical and statistical techniques that can be brought to bear on such data reduction problems The two main methods used in Acuity are Principal Components Analysis PCA and cluster analysis Principal Components Analysis is a data reduction technique It reduces the complexity of a dataset by deriving a small number of variables components from the data The investigator then ex
64. mmarization The techniques used are very different to those used in Affymetrix software so it is worth spending some time explaining them Background Correction The background correction algorithm operates on each array independently Its main function is to determine an estimate of the background noise from the probe measurements and to then subtract that noise from each probe value The background correction algorithm treats the measured probe values as a random variable S which it then decomposes into a signal component X and a noise component Y Both X and Y are assumed to be independent random variables with X being exponentially distributed and Y being normally distributed Once we have determined the parameters of the X and Y distributions from the original data we may estimate the true signal for each probe given the original measured value for each probe In statistical terms we compute the conditional expectation of X given S Normalization The two normalization algorithms provided as part of RMA in Acuity are Quantile Normalization and Cyclic Lowess Normalization Both attempt to normalize the chip data to a common baseline distribution of probe intensities Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 39 Cyclic Lowess Normalization Cyclic Lowess Normalization is an extension of the standard method of normalizing two color microarrays to the case wh
65. n Acuity you can use the cluster order from a Self Organizing Map analysis or scores from a Principal Components Analysis to produce an optimal ordering of the tree Because Principal Components Analysis scores both substances and microarrays you can sort both substance trees and microarray trees with PCA scores Self Organizing Maps in Acuity cluster substances only so if using a SOM to swap branches you can sort the substances tree only For best results with SOMs sort trees with ann x 1 or 1 x n SOM as SOMs of those dimensions produce a single unambiguous order Algorithm Details Hierarchical clustering in Acuity uses a well known algorithm that has been optimized for speed on large data sets In purely theoretical terms hierarchical clustering scales for speed as n where n is the number of rows substances The algorithm proceeds differently depending on the amount of computer memory RAM that is available If there is not enough memory to perform the entire calculation at once the computational task is re cast so that memory is never exceeded In the latter case the calculation is much slower than in the former Chapter 3 24 Analysis Algorithms K Means and K Medians Cluster Analysis K Means and K Medians cluster analyses fall into the category of non hierarchical cluster analysis along with Gene Shaving and Self Organizing Maps K Means clustering partitions the data into a set of mutually exclusive and exh
66. ng group of substances This may be because you have forced a non hierarchical clustering algorithm to find too many clusters Repeat the analysis with fewer clusters and check them again against the principal components Alternatively select the cluster and switch to the Graph tab where you can view all the expression profiles from the cluster together This gives you another view of all the members of a cluster Web Links and Pathways We can view the gluconeogenesis pathway in Acuity by using a web link to connect to a pathway database such as the Kyoto Encyclopedia of Genes and Genomes KEGG Chapter 4 86 Tutorial is one of the crucial gatekeeper genes in the Diauxic study Right click on the Gene column in the Substance Properties pane and select Sort Ascending to sorts the substances alphabetically by gene name Scroll down to and click on its SGD hyperlink on the Web Links tab The page in the SGD database is opened in the Report tab You can also click on the hyperlink in the KEGG column of the Web Links tab to open the EC 5 4 2 1 page from the KEGG database Scroll down here and click the In the Pathway field click the Map00010 link to open the gluconeogenesis pathway Reproducing the Published Results In Figure 5 on page 685 six main expression profiles and key gatekeeper genes are reported We will try to find these profiles and hence discover these genes e find the profiles you nee
67. nt similarity metrics Self organizing maps SOMs with many different similarity metrics Order dendrograms with SOMs PCA K Means and K Medians with many different similarity metrics Gap Statistic to estimate optimal number of K Means and K Medians clusters Principal components analysis Gene Shaving Find similar expression profile with the following similarity metrics Find similar expression profile to user defined profile Variable selection with diagonal linear and quadratic discriminant analysis Chapter 2 8 Introduction Robust Multichip analysis RMA of Affymetrix probe level data Import and display full annotation data in an unlimited number of columns Import and export gene lists Import and export datasets Import chromosome data Substance lists and associated colors Union and intersection of lists Normalization wizard including ratio based normalization wavelength based normalization print tip lowess normalization with options for centering and scaling data normalization to time points and samples Statistics calculated for replicate microarrays including mean median coefficient of variation standard deviation maximum minimum Significance statistics p values calculated by Two Sample Student s t Test or Mann Whitney test and corrections for multiple hypothesis testing by Bonferroni Step Down Bonferroni Hochberg Sidak Step Down Sidak and Benjamini Hochberg methods Multiple
68. nt to the total signal acquired in the other We are assuming that when scanning a particular sample the total signal in each channel is in fact the same In such a case ratio values may be biased towards one channel To minimize this form of variability you should perform preliminary scans to adjust the PMTs so that they are producing roughly the same response in both channels Normalization Methods In the analysis step of your experiment you can improve comparisons across many arrays by normalizing the data from each array One method of normalization is based on the premise that most genes on the array will not be differentially expressed and therefore the arithmetic mean of the ratios from every feature on a given array should be equal to 1 If the mean is not 1 a value is computed which represents the amount by which the ratio data should be scaled such that the mean value returns to 1 This value is the normalization factor Another method is to choose a subset of the features on an image as control features All substances change expression levels under different conditions Normalization control features should be selected based on their consistent Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 33 behavior in all experimental conditions used on your arrays not on their historical use as housekeeping genes in other molecular biology techniques For example the con
69. ntly show interesting patterns in data Similar comments could be made about hierarchical clustering K Medians is the same as K Means except that each cluster is approximated by the median of its members rather than the mean Algorithm Details The algorithm for K Means clustering is based on the original papers by Hartigan et al 1975 1979 and later work by Linde et al 1980 The basic idea of the algorithm is to begin by estimating n initial cluster centroids The Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 25 elements of the data set are then assigned to the cluster with the nearest centroid and the values of the centroids are updated according to the current elements in the cluster The assignment and updating steps are iterated until the centroids only shift by some minimal amount between iterations Variations on the algorithm exist because of the different techniques possible for initializing the centroid values In Acuity initialization is not done explicitly Rather the algorithm begins by computing one centroid of all elements to be clustered as a starting centroid That centroid is then randomly perturbed slightly above and below its value and these two perturbed centroids become the initial values for a K Means iterated procedure for estimating 2 clusters for the data When those 2 cluster centroids have converged they are used to initialize 4 perturbed cl
70. oducts to analyze oligonucleotide arrays according to OGT s patented methods if those arrays have either been purchased from OGT s licensed suppliers or have been made by the customer under a license from OGT Please contact OGT to enquire about a license under OGT s patents at licensing ogt co uk lt mailto licensing ogt co uk gt USE OF THIS INSTRUMENT WITH MICROARRAYS MAY REQUIRE A LICENSE FROM ONE OR MORE THIRD PARTIES THAT HAVE PATENTS RELEVANT TO SUCH USE AXON INSTRUMENTS MOLECULAR DEVICES DOES NOT SUGGEST OR PROMOTE THE USE OF THIS INSTRUMENT IN A MANNER THAT INFRINGES ON THE PATENT RIGHTS OF A THIRD PARTY YOU ARE ENCOURAGED TO EVALUATE WHETHER A LICENSE IS REQUIRED FOR YOUR SPECIFIC APPLICATION OF THIS INSTRUMENT COMPANIES THAT HAVE INTELLECTUAL PROPERTY RIGHTS IN THE POTENTIAL FIELD OF APPLICATION OF THIS INSTRUMENT INCLUDE WITHOUT LIMITATION AFFYMETRIX INC AFFYMETRIX AGILENT AND OXFORD GENE Licensing Notice 94 e Licensing Notice TECHNOLOGY THIS INSTRUMENT HAS NOT BEEN LICENSED OR APPROVED FOR DIAGNOSTIC APPLICATIONS THE USE OF THIS INSTRUMENT IN CONNECTION WITH MICROARRAYS MAY BE WITHIN THE SCOPE OF PATENTS HELD BY AFFYMETRIX TO THE EXTENT THAT AFFYMETRIX PATENT RIGHTS ENCOMPASS THIS INSTRUMENT OR ITS USE AFFYMETRIX HAS GRANTED A LIMITED PATENT LICENSE FOR RESEARCH USE ONLY AND NOT FOR USE IN DIAGNOSTIC PROCEDURES SUCH LICENSE IF APPLICABLE IS LIMITED TO USE OF THIS INSTRUMENT WITH SPOTTED MIC
71. of the probes across all of the chips and all of the probes in the given probe set The summarized expression value of the probe set for chip j is simply This model or any equivalent alternative may be solved by a variety of numerical methods However since outliers are common in Affymetrix data it is best to use statistically robust methods that are relatively insensitive to outliers Median polish and RLM are two examples of robust algorithms for solving this sort of linear model Median Polish Median polish 15 a very robust non parametric algorithm for estimating the u 0 and terms using medians It performs several sweeps over the data in which it progressively refines its estimates of the various terms While fast the non parametric nature of the algorithm means that it potentially loses some information about the original statistical distribution Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 41 Robust Linear Model Robust Linear Model RLM is an iteratively re weighted least squares algorithm for solving the linear model It achieves robustness by selectively down weighting equations with large residuals It is in a sense a compromise between a very robust non parametric algorithm such as median polish and a traditional non robust least squares solution to the linear model It potentially makes better use of the information within the data but at the e
72. og ratios for example may be undefined because of negative ratios e Select Analysis Find Specified Values e Check the Present in at least option and enter 70 for the percentage Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 71 Check the NOT option at the bottom of the dialog Click OK This query finds substances that are not present in at least 70 of microarrays To remove them now select Analysis Remove Selected Rows Finding Differentially Expressed Genes Once we have transformed and cleaned our dataset we are ready to look for differentially expressed genes There are many ways of identifying differentially expressed genes so let s look at a few of them By Fold Change One way of quantifying differential expression in a dataset is to look at genes that have changed by a certain amount on a specified number of arrays This is sometimes called a fold change filter To find genes based on a fold change filter Select Analysis Find Specified Values Make sure that all options are unchecked Assuming you are working with log ratio data check the Absolute value gt option and enter 2 in at least 2 microarrays Click OK This query finds substances that have changed 4 fold 2 fold in log space on at least 2 microarrays You can see their profiles by switching to the Profiles tab To save these genes in a list with your dataset right click on the Da
73. ose replicate arrays then those statistics are displayed here As the Data tab displays already averaged replicates from within arrays these statistics are slightly different to what you would get if you calculated statistics on all replicates on all arrays Statistics are calculated on selected columns so to calculate statistics e Switch to the Data tab e Select the columns on which to calculate statistics by lt Ctrl gt clicking on their titles in the Data tab e Switch to the Statistics tab e Press the lt F5 gt button to calculate statistics To calculate statistics on replicate features within a single array simply change the averaging method that is used on the Data pane You can do this by selecting a method from the list box at the top of the Data pane or by using the Configure Current Data Type to Retrieve dialog box Advanced Tab The Advanced tab displays various advanced statistical data types such as p values principal components scores and correlation coefficients which have been calculated by Acuity statistical analyses The use of the Advanced tab is discussed more below under Performing Analyses Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 59 Views Pane The Views pane contains the many graphical and other derived views of the data that are possible in Acuity Profiles Tab The Profiles tab graphs rows of data for selected substances ag
74. oth plots are M vs A log ratio vs average intensity The lines are print tip lowess curves i e the lowess smoothing curves for the data from each block the data from each block is smoothed separately to account for variation among print tips On the unnormalized data the lowess curves show how much the data will be normalized by the specific lowess method chosen i e for the specific values of smoothing centering and scaling etc to produce the normalized Chapter 3 36 Analysis Algorithms distribution in the bottom pane That is the lowess curves are generated from the lowess normalization options that you choose in the Normalization Wizard On the normalized data the lowess curves show how much residual non uniformity there is in the normalized data again based on the options chosen in the Normalization Wizard In this particular normalization of an array with 16 blocks you can see from the lowess curves that on all blocks low and high intensities are normalized much more strongly than mid range intensities You can also see from the smoothing curves in the bottom pane that there is very little non uniformity left in the data after the smoothing is applied You might see residual non uniformity if your original distribution is very strongly skewed due to for example a large number of control features on your array If more than 10 of features are control spots that have ratio values very different to the rest of your da
75. our of the Acuity interface introducing the many different interface options in Acuity It is worth going through this part of the tutorial at least once so that you know just what is possible in the Acuity interface The second half of the tutorial which begins with the Performing Analyses section the guides you through a sample experiment similar to one of the first time series microarray experiments DeRisi et al 1997 Exploring the Metabolic and Genetic Control of Gene Expression on a Genomic Scale Science 278 680 686 The data from this experiment is installed in the Acuity database as demonstration data This part of the tutorial emphasizes the scientific aspects of using Acuity and highlights the important data transformations and analyses with which you should be familiar Chapter 4 48 e Tutorial This Tutorial should be read together with the following documents e The Acuity Online Help which provides extensive documentation on the controls in every dialog box in Acuity e The Axon Guide to Microarray Analysis which is on the Acuity installation CD and which can be downloaded from the Axon web site There is more than one way to use Acuity This tutorial is designed to introduce you to its major features and to explain their use Equipped with this knowledge you can be confident of exploring the program for yourself Starting Acuity and Connecting To a Database We assume that Acuity has been installed
76. r of immediate advantages the overall structure of the data is revealed and the clustering is much quicker Binary metrics are not very useful for expression studies where one is looking at continuously varying levels of expression They may be useful in studies such as Comparative Genomic Hybridization CGH where one is looking for the presence or absence of genomic DNA Correlation Coefficients Pearson This is the familiar Pearson s correlation coefficient Centered Includes the forced assumption that the mean of a row is zero i e the mean of the row is subtracted from each value Absolute Uses the absolute value of the Pearson correlation When using this method correlated and anti correlated genes are clustered together Chapter 3 18 Analysis Algorithms Spearman s rho Spearman s rho is similar to Pearson s correlation coefficient except that it is calculated on ranks rather than data values That is when calculating the correlation the actual numbers don t matter just their order within the set Kendall s tau Like Spearman s rho Kendall s tau is a rank based correlation coefficient Both Spearman s rho and Kendall s tau are superior to Pearson s correlation coefficient when there are significant outliers in the data Distance Measures Distance measures are based on common measures of physical distance There are different metrics for continuous data and binary data Continuous Da
77. ree has acquired a sub tree This is because whenever we modify a dataset the modification is recorded in the tree so that we have a record of how we have modified it We can always remove dataset modifications by deleting their entries in the Datasets tree and selecting Data Refresh Data to retrieve the original data from the database Chapter 4 70 Tutorial Combine Columns You can use Combine Columns to average or otherwise combine technical replicates Dye swap Columns You can use Dye Swap Columns to create reciprocal ratios from dye swap replicate arrays so that they can be compared with arrays where the dyes are labeled conventionally Sort Columns By default microarrays in a dataset are organized alphabetically You can use Sort Columns to sort them into their experimental order such as time order or by sample type Remove Selected Rows So far we have been discussing dataset transformations Remove Selected Rows is an important command for performing dataset level quality control you can use it to remove whole rows that fail some row specific property like e Fold change across arrays e Percentage of missing values Removing rows is a two step process you need to find the rows first then you need to remove them To find the rows we use Analysis Find Specified Values Let s remove rows that have fewer than 70 present values These can exist in our dataset because spots failed the Query Wizard or l
78. right 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 13 What PCA Shows Principal Components Analysis provides a low dimensional summary of the dataset If you have a dataset that has three columns you can think of the value in each column as being a coordinate in a dimension and so each row has a position in 3 dimensional space Substances close in that 3 dimensional space have similar expression profiles while substances far apart have dissimilar profiles However because most datasets have significantly more than three columns we use PCA to reduce the dimensionality so that the dataset is easier to visualize In almost any dataset some dimensions e g the values of substances on some microarrays contribute less to the overall variance in the sample than other dimensions Biologically we might say that there are some microarrays on which most features are over or under expressed while on other microarrays most spots have very little change in expression One way of thinking about Principal Components Analysis is that it removes the microarrays dimensions on which there is not much happening leaving only the dimensions that contribute most to the variance Furthermore it orders the dimensions from those contributing most to those contributing least to the variance in the dataset The components that are graphed in the PCA Select Components dialog box are the remaining dimensions transformed in a way that m
79. rrays tab of the Project Tree on the left hand side of the Acuity main window Note that you cannot add data to the root of the Microarrays folder The Acuity Interface The Acuity main application is made up of the Project Tree and data windows You organize data in the Project Tree and view data in data windows Common Tasks By default the Common Tasks pane is docked on the left hand side of the Acuity main window It consists of a list of shortcuts to common tasks that you perform in Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 51 Acuity Instead of having to find the tasks in the menus they are organized by category Simply click on the link and the appropriate dialog box is opened for you Importantly the tasks are organized in the order in which they should be performed For example data normalization must occur before any analyses are performed and this order is reflected in the Common Tasks pane You can show and hide the Common Tasks pane with the View Common Tasks command Project Tree The Project Tree is docked next to the Common Tasks pane on the left hand side of the Acuity main window It consists of three tabs organizing three types of data in the database e The Microarrays tab lists all the microarrays e g GPR files that have been imported to the Acuity database e The Datasets tab lists all datasets sets of spots and analysis results e g
80. s selecting Split Substance Table returns the pane to its original configuration Data Tab The Data tab in the Table displays a single data type from the current data source i e a single GPR column from each microarray Important Note Replicates within microarrays i e spots with the same IDs are automatically averaged in Acuity To see individual feature values for replicate spots use the Features tab in the bottom Views pane You can change the averaging method from the default mean by selecting a method from the list box at the top of the window The display of columns in the Data table is highly configurable e hide data values and display colored cells only use the Data Columns AutoFit Color command e change the color scheme used in the Table use the Configure Color Map command Chapter 4 56 Tutorial e To remove all color from data cells in the Table use the Data Color Map command e To change column widths use the AutoFit commands in the Data Columns sub menu or use the various AutoFit commands in the right mouse menu e To sort data values in a column use the Data Sort Ascending and Data Sort Descending commands or double click on a column title e group together discontinuously selected rows in the table use the Data Group Selection command Selections in the table are always linked with selections in the other panes so selecting in one pane selects in all pan
81. sophila C elegans C Briggsae S cerevisiae SARS In addition there are a number of scripts on the Acuity Report tab that generate CDT files for other organisms Performing Analyses In this part of the tutorial we will walk through the analysis of a typical microarray experiment Not all steps outlined in the tutorial will be appropriate to every microarray experiment but you should be able to learn enough to be able to understand the issues involved in microarray analysis We begin from the assumption that you have already imported GPR files The Acuity installer CD contains two sets of sample GPR files in the Sample Data directory The files in the Diauxic folder are from one of the first published time series microarray experiments DeRisi et al 1997 Exploring the Metabolic and Genetic Control of Gene Expression on a Genomic Scale Science 278 680 686 The purpose of this Chapter 4 64 Tutorial part of the tutorial is to reproduce some of the results in this study which examines changes in yeast metabolism from fermentation to respiration The data from these files is already in the Acuity demo database However you may wish to import them again To import the GPR files in the Diauxic folder e Select File Import Microarrays e Navigate to the folder containing the Diauxic GPR files and select them all e Click Open e Inthe Select Destination dialog box click the Create New Folder icon to create a new
82. sure that you have a dataset open in the active data window e Select Clustering Self Organizing Maps e Click OK The Cluster Progress dialog box is displayed reporting the progress of the task When the progress reaches 100 the clustering result is added to the Project Tree under the dataset that was clustered Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 83 Double click on the result in the Project Tree to display the result in the Visualizations tab Using Self Organizing Maps The result of a Self Organizing Maps cluster analysis is displayed in the Visualizations tab Each cluster is represented as a thumbnail consisting of e A compressed color table containing the colored profiles of all the substances in the cluster e An average trace of the expression profiles of all the substances in the cluster e title bar containing the number of substances in each cluster The grayscale shade of the title bar represents the relative number of substances in each cluster where white is the cluster with the most substances and black is the cluster with the fewest substances The clusters are arranged on a grid according to their similarity to each other similar clusters are close together while dissimilar clusters are separated Clusters diagonally opposite on the grid are essentially anti correlated To show the color table only without the average profile select Color M
83. ta Euclidean Squared dij Xin City Block Manhattan dij Kix The City Block measure and the Euclidean measure give similar results but the City Block is less affected by extreme outliers as the values are not squared Canberra 1 0 when xir x 0 dij otherwise The Canberra metric is self standardizing This metric can be unstable when there are many values near zero which happens with log transformed array data It usually applies to non negative data Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms e 19 Bray Curtis 4 0 when xix 0 dij LE xix xix Xi otherwise Bray Curtis is usually applied to non negative data only Soergel di 0 when xix 0 dij MAX Xix otherwise The Soergel metric applies to on negative data For binary data 0 1 values it gives the same result as the Jaccard Metric Binary Data Metrics for binary data are generated by considering the table of co occurrences of two substances over a set of p arrays Substance pew poo Substance j From this table we can see that on a of the p arrays both substance i and substance j are positive for d of the p arrays both substances are negative for c of the arrays substance i is positive and substance j is negative and on b of the arrays substance j is negative
84. ta you may have to adjust the number of iterations of the lowess smoother Lowess normalization is treated differently by the Acuity database than ratio normalization When you do a linear ratio normalization all data types for the microarray in the database that can be normalized are normalized When you do a lowess normalization because each data point on the array has to be individually changed Acuity creates two new data types in the database A and M A is an average intensity constructed from the two intensity values that you select while M is the lowess normalized log ratio To use lowess normalized data in your downstream analyses use the M data type Lowess normalization is described fully in Dudoit et al 2002 and Yang et al 2002 Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 37 T ip o wo 6 8 10 12 14 F635 Mean F532 Mean Normalized Ratio T G 2 ip wo 6 8 10 12 14 F635 Mean F532 Mean Figure 2 Lowess normalization from the Acuity Normalization Viewer Chapter 3 38 Analysis Algorithms Robust Multichip Analysis RMA Robust Multichip Analysis is a method of taking Affymetrix probe level signal intensities and performing the following operations e Background Correction e Normalization e Su
85. ta tab and select Create Dataset Quicklist Your dataset acquires another tree this time of quicklists and the quicklist is saved there Chapter 4 72 Tutorial By Statistical Significance The fold change method of finding differentially expressed gene is a rather blunt instrument One adjusts the value of the fold change and the number of microarrays until one has a manageable number of genes A more objective way of quantifying differential expression in a dataset is to look at genes that have statistically significant differences in expression among groups of arrays One statistical test we can do for two groups is a two sample t Test Select Advanced One and Two Sample Significance Tests Select the first test Student s t Test equal variances Click OK to move to the next step In the next dialog box you have to specify the microarrays in each group In the Diauxic demo data in the Acuity database even though it is a time course there are two groups of arrays the first five and the last two We can create two groups of microarrays by clicking the Create from Microarrays button Select the microarrays in the first group and create a group by moving them to the list on the right hand side Do the second for the second group Click OK to move to the next step The next dialog reports the results of the t test as a sorted list of p values You can save all p values to the Advanced tab by clicking the Save button Yo
86. thms the points in each cluster Minimizing the Euclidean Squared distance of the cluster s points to the cluster s centroid naturally gives the centroid as the arithmetic average For K Medians Analysis the cluster centroids are the medians of the points in each cluster Minimizing the City Block distance of the cluster s points to the cluster s centroid naturally gives the centroids as the median in this case Gap Statistic Analysis The Gap Statistic proposed by Tibshirani et al 2000 is method for estimating the number of clusters in a set of data It can use the output of any clustering algorithm but in Acuity we restrict it to K Means and K Medians The Gap Statistic compares the change in the within cluster dispersion to that expected under reference null distribution Tibshirani et al offers two options for generating the reference distribution e Generate each reference feature uniformly over the range of the observed values for that feature e Generate the reference features from a uniform distribution over a box aligned with the principal components of the data The implementation of the Gap Statistic in Acuity uses b as the reference distribution as this method takes into account the shape of the data distribution Method b assumes that the sample data is column centered so we have the requirement in our implementation that the data is column centered After column centering a large number of data poin
87. transformations are popular because if we do a scatter plot of red intensity versus green intensity we often see the lower part of the scatter plot curving towards the red or the green when we expect a straight line through the origin A linear transformation shifts the distribution up or down without changing its shape a non linear transformation changes the shape of the distribution One of the more common non linear normalization methods used on microarray data is lowess locally weighted scatter plot smoothing What imbalances in the data does Lowess normalization correct for Commonly cited defects include the properties of the different dyes used e g different labeling efficiencies and Chapter 3 34 Analysis Algorithms scanning properties and experimental variability resulting for example from separate reverse transcription and labeling of the two samples Lowess normalization is somewhat problematic because the defects in experimental design or execution that are being corrected are not sufficiently well understood There is no mathematical model of the properties of the dyes or their labeling efficiencies analogous to the mathematical model of the response of PMTs at different intensities Therefore we recommend caution when using lowess normalization Whether or not you use lowess normalization will depend on your attitude towards the use of statistical techniques for data correction Statistical techniques like
88. trol spots might be such that each is expected to have a ratio of 1 Hence the mean of the control features should be 1 Assuming that variations are uniform across the array a single normalization factor can be calculated from these features and then applied to the whole array Both of these methods are linear and ratio based that is they correct every feature on the array by the same multiplicative factor and they correct intensities in order to balance ratio values The most common reason for normalizing microarray experiments is to correct for a scanner with an uncalibrated ratio channel For a data distribution in which the average ratio value is different from 1 0 we can scale the intensity data in each channel with a linear transformation so that the ratio is equal to 1 0 Since PMT response is linear over a wide range of incident light this type of data correction is equivalent to performing the experiment again with the PMTs calibrated The linear transformation matches the instrument adjustments and so we are justified in correcting the data A linear transformation to correct the balance between red and green across a whole slide is one method of normalization There are a number of non linear transformations that are also used to correct microarray data A non linear transformation corrects different spot intensities differently so that for example low intensity features are shifted differently to high intensity features These
89. ts are negative and all of the binary K Means and K Medians metrics are incompatible with negative values Therefore you cannot use the Bray Curtis Jaccard Simple Matching or Soergel metrics with the Gap Statistic Apart from choice b for the reference distribution the implementation in Acuity follows the computation as specified on page 6 of Tibshirani et al Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms e 27 It is worth emphasizing that the optimal cluster size determined by the Gap Statistic is not the cluster with the largest Gap value Rather it is the smallest cluster whose Gap value is closer than one standard error to the Gap value of the next cluster More formally for a cluster size k and standard error 5 in the reference distribution the optimal cluster size is smallest k such that Gap k gt Gap k 1 5 The standard error 5 is displayed on Gap Statistic graphs in Acuity as error bars Self Organizing Maps Analysis Self Organizing Maps falls into the category of non hierarchical cluster analysis along with K Means K Medians Gene Shaving Self Organizing Maps analysis is used to group substances with a similar pattern over arrays together The inputs and outputs for a Self Organizing Maps analysis are the same as for a K Means or K Medians cluster analysis Self Organizing Maps are a simple amendment to K Means analysis The key addition to K Means is
90. u can select all genes that pass a p value threshold say 0 001 by entering 0 001 in the Select all substances field and clicking select Save these substances as a dataset quicklist by selecting Create Dataset Quicklist from the right mouse menu Click Close Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 73 You can view the expression profiles of the substances that you selected by switching to the Profiles tab If you have more than two groups in your experiments for example from multiple treatments you can use Advanced One way ANOVA For Multiple Groups to find the differentially expressed genes It works the same as the t Test but on multiple groups One can look at the genes that are common between the fold change filter and the t Test e Select both quicklists that you created by holding down the lt Shift gt key while selecting them in the tree e Select Create Intersection Quicklist from the right mouse menu The intersection quicklist contains substances that are common to both quicklists To see their profiles e Right click on the intersection quicklist e Select Apply as Selection e Switch to the Profiles tab Another way of looking at the interaction between fold change and statistical significance is to plot log ratio against p value or log p First let s convert our p values to log p e Assuming that you saved your p values to the Advanced ta
91. uickly for example to see which microarrays have the most missing values e Ensure that you have a dataset open in the active data window e Select Clustering Hierarchical Clustering e Under Data to Process select None Color Map Only e Click OK Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Tutorial 81 This feature can be particularly useful when used together with the described below Match Expression If you want to find substances that have a similar expression profile to a selected substance in a dataset you can perform a quick cluster of a data source using Advanced Match Expression on Mean This command sorts the data source by their similarity to the selected substance To use Match Expression Select a substance row in the Table View in which you are interested Select Advanced Match Expression on Mean The selected substance is now in the first row of the table and the rest of the table is sorted from most correlated to the selected row to least correlated to the selected row You can save the correlation coefficients to the Advanced tab by clicking the Save button After doing a Match Expression you can select the first 10 or 20 substances in the table and visualize them in a number of different ways Switch to the Graph tab to see them graphed together Select Average Mode from the right mouse menu to see an average trace of the selected su
92. uster centroids and so on doubling the number of centroids at each step This initialization process is called splitting When the number of clusters desired is not a power of 2 the final splitting step involves using the only a subset of perturbed centroids from the previous step By employing this splitting method the initialization of the centroids at each split will be much more likely to provide good starting values for converging to a good local optimum close to the global optimum as opposed to a poor local optimum The latter may occur if the centroids are initialized in a more random manner However because the initial perturbations are random K Means cluster solutions are not entirely reproducible That is running the same analysis on the same dataset twice produces slightly different results This is not an error or a drawback of the algorithm Rather it demonstrates the fundamental arbitrariness of any cluster solution Cluster membership is always affected by assumptions that one makes in the implementation The K Means and K Medians algorithms in Acuity 4 0 use a similar set of metrics as are available for hierarchical clustering For more information on the metrics see their description above in the Hierarchical Clustering section For K Means Analysis the Euclidean Squared metric is usually the most appropriate as for that metric the cluster centroids are arithmetic averages of Chapter 3 26 Analysis Algori
93. uter on which SQL Server is installed and you are changing the password from that computer To do this e Click the Change Password button in the Welcome to Acuity login dialog box e Type a new password in the Change Password dialog box Importing Microarray Data Our first task is to populate the empty database by importing microarray data Acuity can import any tab delimited text file that has one row of column titles and one column labeled ID that contains the unique identifier of each substance However using the GPR file format has a number of advantages such as enabling the import of GenePix Results JPG images Chapter 4 50 Tutorial To import GPR files Select the File Import Microarrays command Navigate to a directory that contains GPR files There are some sample GPR files on the Acuity installer CD in the Sample Data directory You may like to import the GPR files in the Diauxic sub directory as we will use these later in the sample experiment Select more than one file by holding down the Ctrl key when selecting Click Open The Import Microarrays dialog box is displayed Select a folder on the Microarrays tab in the Select location pane You can create new folders and rename them or existing folders with the New Folder and Rename buttons If there are Results JPGs associated with your GPR files they are imported automatically Click OK Once imported microarrays are listed on the Microa
94. visualization as PDF BMP WMF Export animated 3D scatter plots as AVI Database Support for Microsoft SQL Server 2000 and Oracle 9 ODBC compliant Full client server model for effortless local LAN or remote TCP IP access Tools for creating and managing users and groups Users with read only read write or lab head permissions Advanced database search tools Advanced database management tools such as a database optimizer to re build database table indices Organize substance annotations into warehouses and genomes True copy and paste of microarrays in the database Attach import any file type to a microarray or a dataset in the database Database backup and restore utility SQL Server only Database Recycle Bin to permanently delete or restore deleted data Compact database tool to minimize database size on disk Universal text file import including GenePix Pro 5 0 GPR files and Affymetrix CEL CHP and CDF files Includes Microsoft SQL Server 2000 Desktop Edition Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 11 Chapter 3 Analysis Algorithms Theory and Use Acuity employs a number of advanced algorithms for microarray analysis This chapter explains the uses of these analyses their limitations and how to interpret their results The Fundamental Assumption In microarray data analysis more specifically in time course experiments we
95. w in the Performing Analyses section Quicklists Tab A quicklist is a set of substance and microarray names It is a way of keeping lists of substances handy so that you can find them quickly in any microarray or dataset That is at any time the substances or microarrays in a quicklist can be selected in the Acuity interface There are two types of quicklists in Acuity e Global quicklists which are organized in the Quicklists tab and which can be applied to any microarray or dataset Chapter 4 54 Tutorial e Dataset quicklists which are saved with a dataset in the Datasets tab To create a global quicklist e Select some rows from the Table view of a microarray or dataset e Select the Create Global Quicklist command from the right mouse menu e Give the quicklist a name and click OK To highlight all the substances from a global quicklist in the current dataset e Open the dataset by selecting it on the Datasets tab and choosing Open Selected from the right mouse menu e Select the quicklist from the Quicklists tab and choose Apply As Selection from the right mouse menu e The substances in the quicklist are highlighted in the data window Quicklists are saved with colors associated with them so that you can add a color to substances throughout the Acuity interface This is particularly helpful when tracking substances from multiple quicklists To apply a quicklist color to substances in the current data window
96. w about a dataset Furthermore the various similarity metrics and linkage methods introduce different assumptions to the process so it is worth trying a number of methods just to see the results Chapter 4 80 Tutorial Branch Swapping Dendrograms with PCA or SOMs As explained above when doing a Principal Components Analysis each gene is given a score for each component Acuity can use the resulting order to apply an optimal branch order to a dendrogram thereby partially obviating the need to swap branches manually To do this first perform a PCA by selecting Clustering Principal Components Analysis Once it is finished open the Clustering Hierarchical Clustering dialog box In the Order substance branches by field the recently completed PCA is listed along with any SOM that has been performed on the dataset Select the PCA from the list and the component to use then click OK to start the hierarchical cluster Because Self Organizing Maps SOM order their clusters on a 2 dimensional grid one can use this ordering to swap branches To use this feature first perform a Self Organizing Map analysis of the dataset Typically we perform a 1 x n orn x 1 SOM as we are interested in ordering the tree in one dimension only Color Map Only Another feature of hierarchical clustering is that you can create a completely unclustered color map of a whole dataset This allows you to view the global structure of a dataset very q
97. xpense of significantly greater computational effort Algorithm Complexity Large datasets can be fairly time consuming to run through RMA In order to provide some guidance as to what to expect we give the approximate order of complexity of each algorithm Algorithmic complexity is estimated in terms of the number of arithmetic operations the algorithm must perform for example multiplication of two numbers constitutes a single operation For some algorithms the exact behavior as a function of the number of arrays and number of probes is impossible to precisely characterize the complexity quoted in the table below should be taken as a guide only In the table below the main variables are e the number of Affymetrix chips being processed as a batch e the number of Perfect Match probes single Affymetrix chip The big O notation means that the dominant scaling factor in the complexity of the algorithm is of the order of the item in the parentheses For example O P implies that the number of operations performed by the algorithm scales quadratically in the number of arrays P Chapter 3 42 e Analysis Algorithms Algorithm Complexity Background Correction O PNlogeN Quantile Normalization O PNlogeN Cyclic Lowess Normalization O P NlogeN Median Polish Summarization O PNlogeN RLM Summarization O P N References General Multivariate Statistics Hartigan J A Clustering al
98. xperiment are already intrinsically ordered In an experiment where each array corresponds to a tissue sample from a different patient for example you would cluster on both substances and arrays With any method that reduces the dimension of the data however finer structure can be lost For example suppose the expression of some subset of genes divides the samples in an informative way correlating with the rate of proliferation of tumor cells for example whereas another subset of genes divides the samples a different way representing the immune response for example Then methods such as two way hierarchical clustering which seek a single reordering of the samples for all genes cannot find such structure Acuity 4 0 User s Guide Copyright 2005 Axon Instruments Molecular Devices Corp Analysis Algorithms 23 Optimizing Branch Swapping in Dendrograms with SOMs and Principal Components For tree containing n items there 25 different ways of ordering the tree that are consistent with the results of the clustering algorithm Some orderings of trees reveal the tree structure better than other orderings For this reason you can manually swap the branches under a node by selecting the node then selecting the Swap Branches command from the right mouse menu on the dendrogram Manual branch swapping is a trial and error process that is extremely time consuming on large dendrograms it is practically impossible However i
99. y not be copied for resale or for bundling with other products without prior written permission from MOLECULAR DEVICES Restrictions on Use and Transfer You may not reverse engineer decompile disassemble or create derivative works from the SOFTWARE Export of Software You agree not to export the SOFTWARE in violation of any United States statute or regulation Ownership of Software and Media CD ROM You own the media CD ROM on which the SOFTWARE is recorded but MOLECULAR DEVICES owns the SOFTWARE and all copies of the SOFTWARE Product Improvements MOLECULAR DEVICES reserves the right to make corrections or improvements to the SOFTWARE and its documentation and to the related media at any time without notice and with no responsibility to provide these changes to purchasers of earlier versions of such products Customer Licensing Agreement 92 Customer License Agreement Term This license is effective until terminated You may terminate it by destroying the SOFTWARE and its documentation and all copies thereof This License will also terminate if you fail to comply with any term or condition of this Agreement You agree upon such termination to destroy all copies of the SOFTWARE and its documentation Limited Warranty and Disclaimer of Liability MOLECULAR DEVICES warrants that the media on which the SOFTWARE is recorded and the documentation provided with the SOFTWARE are free from defects in materials and workmanship und

Download Pdf Manuals

Related Search

Related Contents

guide - PLATEFORME OUVRAGES D`ART - "PILES"  MANUAL DE USO E MANUTENÇÃO DUCATO  ISTRUZIONI D`USO - INSTRUCTIONS  取扱説明書 - Panasonic  Trident Perseus  Descargar / Cerabar T PMP131 - Endress+Hauser E  Descarga - refrigeracion guevara.com.mx  Avaya Multimedia Contact Center User's Manual  アンパンマン-ベビー用品全般 電子レンジでの温めはできますか  35, 50, 70 et 100 - Akoestix Products & Projects  

Copyright © All rights reserved. Failed to retrieve file