User manual v1.0
Contents
1. control no STN GP DA ablation no STN GP DA DA lesion no STN GP DA cortical ablation and DA le no no no no no no Simulation file single_runl single_runlb single_run2 single_run3a single_run3b single_run4a single_run4b single_rundc single_run4d single_run5_la single_run5_1b single_run5_Ic single_run5_1d single_run5_2a single_run5_2b single_run5_2c single_run5_2d single_run5_3a single_run5_3b single_run5_3c single_run5_3d single_run5 4a single_run5_4b single_run5_4c single_run5_4d single_run5_5a single_run5_5b single_run5_5c single_run5_5d single_run5_6d single_run6a single_run6b single_run6c Parameter file pars parslb pars2 pars3a pars3b pars4a pars4b pars4c pars4d parso_la pars5_1b parso_Ic pars5_1d pars _2a pars5_2b parsd_2c pars5_2d parsd_3a pars5_3b pars5 3c pars5_3d parso_4a pars5_4b pars5 4c pars5_4d parso_5a pars5_5b pars5_5c pars5_5d pars5_6d pars6a pars6b pars6c Table 1 Table of files corresponding to each experimental study in Humphries et al 2006 DA dopamine LFO low frequency oscillation Analysis flag file sum_flags sum_flags sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_flags45 sum_fl2. same number of cells per model as were sampled on average per animal For example in the control condition of Magill et al 2001 slow wave oscillation study they used 7 rats sampling a total of 28 subthalamic nucleus STN cells and 18 globus pallidus GP cells To emulate this condition we therefore ran 7 models all with the same parameter set sampling 4 cells from each STN giving a total of 28 model STN neurons and 3 cells from each GP giving a total of 21 model GP neurons We term this batch of simulations a virtual experiment We run multiple batches usually 50 to better understand the spread of results predicted by the model All of this process the multiple simulations sampling of cells batches and so on is automated by models_as_individuals and the functions it calls in addition to those already listed extract_spikes performs the sampling operation other functions are discussed below Each simulated experimental condition is run by a top level function single_runX which calls models_as_individuals with the parameters necessary for emulating that experiment s struc ture e number of batches to run e number of models in a batch single virtual experiment e which structures to sample from e the number of cells to sample from each structure in each model e the extraction threshold not used in Humphries et al 2006 This sets a threshold level in spikes s for the minimum mean firing rate a cell m
3. that it knows which analyses to combine Finally there are a set of post processing scripts in directory Postproc which take the final filename list for all batches and automate comparisons of the analysed results across ex perimental conditions These scripts save much of their output into ASCII format files suitable for importing into graphing programs References Humphries M D and Gurney K 2007 A means to an end validating models by fitting experimental data Neurocomputing in press DOI 10 1016 j neucom 2006 10 061 Humphries M D Stewart R D and Gurney K N 2006 A physiologically plausible model of action selection and oscillatory activity in the basal ganglia J Neurosci 26 12921 12942 Jarvis M R and Mitra P P 2001 Sampling properties of the spectrum and coherency of sequences of action potentials Neural Comput 13 4 717 749 Magill P J Bolam J P and Bevan M D 2001 Dopamine regulates the impact of the cerebral cortex on the subthalamic nucleus globus pallidus network Neuroscience 106 313 330 Marsaglia G and Tsang W W 2000 The ziggurat method for generating random variables J Stat Soft 5 8
4. User Manual for the Spiking Neuron Basal Ganglia Model Version 1 0 14 12 2006 Mark D Humphries m d humphries sheffield ac uk Abstract The spiking neuron model of the basal ganglia was developed over a period of 6 years and the resulting code base reflects the complexity of this process These notes are intended to guide the end user through the basic structure and use of the code Please e mail any suggestions requests or queries for incorporation into future revisions 1 Introduction The following is a short user manual for the spiking neuron basal ganglia BG model guiding the user through the basic steps of calling the correct functions and how the results presented in the Journal of Neuroscience paper Humphries et al 2006 were analysed It is anticipated that the copious Help sections and comments throughout the code will be sufficient to understand the detail All names in bold are MATLAB functions or scripts All custom MATLAB functions have full Help comments describing their function calls The code is provided in exactly the form in which we used it and so there are some inevitable historical oddities comments not matching text redundant parameters etc Nevertheless by doing this we can ensure that our results are replicable Unpack both ZIP files to a single directory then add that directory and all its sub directories to your MATLAB path 2 Basic code The central function is BATCH_BG_heterogenous_AMPA_NMDA w
5. ags45 sum_flags45 sum_flags45 sum_flags45 sum_flags6 sum_flags6 sum_flags6 gt gt idx find in_n X array index of all events for that neuron gt gt times in_t idx dt corresponding time steps converted to seconds We have also included the Chronux toolbox Jarvis and Mitra 2001 that we used as a separate ZIP file The latest version can be downloaded from chronux org if required 5 2 A simulation batch The models_as_individuals function deletes all the individual output files that each call to BATCH_BG_heterogenous_AMPA_ NMDA creates after the appropriate spike trains has been extracted It then calls a user specified function to analyse each individual spike train for the published work this was batch_analyse_stngpe A further user specified function for the published work this was combine _stngpe_analysis is called at the end of each batch to pool all of the analyses from the batch into a single file The output filenames from each of these two analysis processes are stored in a single file covering all batches The batch_analyse_stngpe function we supply contains a fairly exhaustive list of analyses invoking many of the individual spike train analysis functions in the supplied MATLAB toolbox A user defined analysis flag file determines which of these analyses is actually carried out for a given type of experiment see e g sum flags The function that combines the analyses also makes use of this flag file so
6. analytically solvable model neurons between the flexibility of a time driven system and the efficiency of an event driven system The C code was written for optimal performance and as such is not particularly human readable We made full use of the analytically solvable neuron model by moving all constants outside of the neuron update loop they are computed in BATCH _BG_heterogenous AMPA NMDA before the simulation engine is called This enabled us to achieve faster than real time perfor mance i e 5 seconds of simulation time took less than 5 seconds of computation on a modest desktop PC Intel P4 2 4 GHz for models with thousands of neurons Also note that the random number generator is given in full in the code based on the modified Ziggurat method of Marsaglia and Tsang 2000 If extensions to this code are required such as a new neuron model please contact us and we would be happy to help modify the simulation engine 3 Code for structuring the simulations Much of the code base is devoted to the structure of the simulations The central function is models_as_individuals which sets up our models as animals simulation protocol see Humphries and Gurney 2007 for why we think this is a good idea The basic idea is briefly explained on p 12925 of the Journal of Neuroscience paper when replicating a particular con dition of an experimental study we ran as many simulations as there were animals in that condition and sample the
7. and 64 bit Linux but compiling a MEX file for any system is straight forward 2 1 The parameters file A full parameters file is typified by the pars script The most altered parameters appear at the top of the file including global dopamine level simulation time and input type and rates After this is a section which specifies the global network parameters number of neurons per channel and so on followed by parameters for each class of neuron These are followed by weights and transmission delays coefficients for the STN and GP dopamine models shunting inhibition parameters intrinsic currents driving spontaneous and burst firing and the option of including injection currents All parameter values are given in SI units seconds amps volts etc where required This avoids the inevitable errors that occur in the worryingly common practice of working in phys iological units ms pA mV etc 2 2 Undocumented parameters There are some parameters in the code that are not mentioned in Humphries et al 2006 because they were not used However they may be useful in future experiments and so we briefly describe them here 1 trace_n the neuron given here e g GPe 75 is the 75th GPe neuron has complete traces of some of its variables saved in the results file from BATCH_BG_heterogenous_AMPA_ NMDA The current list is distal inhibitory synaptic input I a membrane potential V and total gating variable Q The variables th
8. at are saved can be edited in GHS_LIF_solver_shunt_AMPA_ NMDA and a commented out list of suggestions is included in that code don t forget to re compile after editing 2 refresets the membrane potential to this value after a spike is fired This allows for explicit after hyperpolarisation to be modeled if required though does introduce small numerical errors as the analytic solution to V does not account for this 3 STN_ext_ratio will scale the cortical subthalamic weight to be this fraction of the cortical striatal weight 4 PSP_sigma standard deviation of normally distributed noise for post synaptic potential size Allows for separate parameterisation of synaptic noise due to e g spike failure or graded vesicle release from membrane noise 5 Injection current end of script the simulation code contains the ability to inject a current pulse train of specified width and frequency into user selected cells This allows for simple replications of current injection studies in intact BG 2 3 The simulation engine Our central simulation engine GHS_LIF_solver_shunt_AMPA_NMDA was written in C to gain a vast performance increase over native MATLAB code It uses a mixed time driven event driven system where each neuron s membrane potential is updated every time step but its synaptic input is only processed when an pre synaptic event occurs and accounting for the transmission delay We feel this achieves a good trade off for
9. ce We consider the file pars as the baseline state of the model being the nominal quiescent in vivo state of the BG Table 1 lists the single_runX and parameter files that correspond to each published study in Humphries et al 2006 To run an existing experiment with the model do the following 1 edit pathroot variable in the appropriate single_runX function so that it points to a suitable directory for the storage of the output files 2 invoke that function at the command line in MATLAB Use these single_runX and parameter files as templates for further experiments with the model For example to look at the effects of a systemic D1 agonist on the y band oscillations copy and rename single_run6a and pars6a Edit the renamed single_runX function to include the new name for the parameter file a new experiment name and an appropriate path edit the parameter file setting dop1 1 Then invoke the new single_runX function from the MATLAB command line 5 Analysing the results 5 1 A single simulation A single run of BATCH_BG_heterogenous_AMPA NMDA results in a single output file in MAT format which contains most of the important simulation parameters to allow future reconstruction of the simulation and the spike data in a compressed format Output spikes are encoded in two arrays the first in_t contains the time step indices of all spikes in the order they occurred the second in_n contains the corresponding index of the neuro
10. hich accepts five ar guments including a string containing the filename of the parameters file Every user definable parameter is contained in this file so that BATCH_BG_heterogenous_AMPA NMDA only requires editing if new structures e g thalamus cortex are added to the model A sec ond version BATCH_BG_heterogenous_AMPA_NMDA2 has the additional flexibility of assigning separate weights to the AMPA and NMDA receptors though this was only used in Humphries et al 2006 to study the effects of NMDA receptor blockade on the y band activity These functions call in turn 1 BG_GHS_input which generates the cortical input streams according to the values set in the parameters file 2 BG_net_shunt_AMPA NMDA which creates the network specified in the parameters file The random number generator is reset before this function is called to ensure that the same network is specified each time for a given parameter set In turn this function calls e PSPtoPSC which converts the post synaptic potential size give in volts within the parameters file into the appropriate current step that should be elicited by the arrival of a spike at the pre synaptic membrane given the post synaptic neuron s average membrane properties 3 GHS_LIF solver shunt AMPA NMDA which simulates the specified network for the specified time period given the specified inputs This is a MEX file compiled from the original C Compiled versions exist for Windows XP Linux
11. n that emitted the spike Similarly the simulated cortical input spike trains are encoded in the corresponding out_t and out_n arrays My MATLAB toolbox of analysis functions is provided LIF tools Analysis Each has a detailed Help section Each typically requires a single neuron s time series as its first argument For neuron X this can be extracted in MATLAB from the spike encoding arrays thus For historical reasons the actual values in input_array are scaled down by a constant factor specified in the pars file and it is the scaled values reported in the paper Experiment Tonic rates collaterals Tonic rates no collaterals Selection and switching Selection and switching low DA Selection and switching high DA LFO LFO LFO LFO sion LFO LFO LFO LFO ablation and DA lesion STN GP DA LFO control no STN DA LFO ablation no STN DA LFO DA lesion no STN DA LFO ablation and DA lesion STN DA LFO control no GP DA LFO ablation no GP DA LFO DA lesion no GP DA LFO ablation and DA lesion GP DA LFO control no collaterals LFO ablation no collaterals LFO DA lesion no collaterals LFO ablation and DA lesion collaterals LFO control no urethane LFO ablation no urethane LFO DA lesion no urethane LFO ablation and DA lesion urethane LFO ablation and DA lesion Ca t in STN y band control y band D2 agonist y band NMDA blocker in GP control cortical ablation DA lesion
12. ust have to be considered for sampling The effects of systematic sampling bias in micro electrode recordings can be emulated by setting this to a positive value and with e the file path for saving all output files e the name of the parameter file associated with that experiment e the name of the analysis flag file for that experiment see below e an identifying name for that experiment used as prefix for the output files e the type of analysis required if any see below 3 1 Selection and switching experiments These are handled differently to the rest of the model experiments as they require running a large batch of simulations on a single model If models_as_individuals is invoked with the type parameter set to the string sel then it in turn calls batch_selection_grid_DA This function runs a complete batch on a single model one simulation per row in the array input_array which is loaded from the MAT file input_grid The results of the batch are each classified as no selection selection switching dual selection or interference by the function mean_output The selection threshold 6 is specified within the batch_selection_grid_DA function 4 Running the simulations For simple simulations BATCH BG heterogenous AMPA NMDA can be invoked di rectly from the MATLAB command line with appropriate arguments see its Help comments This will require a suitable parameter file for which any of the supplied set will suffi
Download Pdf Manuals
Related Search