Poisson-GLMs-for-Neural-Spike-Train-Analysis

This repository provides a Python implementation for analyzing multi-neuronal spike trains with Poisson Generalized Linear Models.

1. The dataset used in this repository can be downloaded here:

   Tim Blanche. Multi-neuron recordings in primary visual cortex:

   http://crcns.org/data-sets/vc/pvc-3,3702016.

2. For a formal description of the Poisson-GLM framework demonstrated in this repository, you may refer to:

   Truccolo, W., Eden, U.T., Fellows, M.R., Donoghue, J.P. and Brown, E.N., 2005. A point process framework for relating neural spiking activity to spiking history, neural ensemble, and extrinsic covariate effects. Journal of neurophysiology, 93(2), pp.1074-1089

   https://www.ncbi.nlm.nih.gov/pubmed/15356183

Setting-up the tutorial

1. Clone the repo

   git clone https://github.com/ifiaposto/Poisson-GLMs-for-Neural-Spike-Train-Analysis.git
   

2. Install the requirements

   pip install -r requirements.txt
   

Running the tutorial

Discretization of the neuronal data

This step discretizes the spike trains of the neurons: It counts the number of spikes that occured in each time bin of the discretized time interval.

python3 neuron_count_data.py <nof neurons> <time bin size>

Example:

python3 neuron_count_data.py 25 1

Create the training and testing dataset

This file creates the training and testing files to be used in the regression: It finds the last time bin so that a total number of 'nof_spikes' spikes across all neurons has occured up to that time bin (included). Subsequently, it finds the next time bin which contains the next 'nof_spikes' spikes (immediately following the spike train used for training). The training and testing files per neuron which contain the spike counts up to these time-bins are created.

python3 neuron_count_data.py <nof neurons> <time bin size> <nof_spikes>

Example:

python3 neuron_data_crop_train_test.py 25 1 4000

Create regression covariates

This file prepares auxiliary files needed for the poisson regression:

It creates the covariates of the regression by merging the counts of spikes of all neurons that happened 'Q' time bins in the past, where Q is the degree of regression.

The dependent variable is the spike counts in each time bin.

python3 neuron_count_data.py <nof neurons> <time bin size> <nof_spikes> <degree of regression>

Example:

python3 neuron_mutual_regression.py 25 1 4000 1

Fit the Poisson-GLM

This step fits a Poisson-GLM for spike trains. In the results directory, it prints the training and testing discrete-time loglikelihood achieved and its continuous-time approximation (see the paper cited above). It also returns a csv/ neuron that contains the learned parameters.

python3 fit_spike_trains_mutual_regression.py <nof neurons> <time bin size> <nof_spikes> <degree of regression>

Example:

python3 fit_spike_trains_mutual_regression.py 25 1 4000 1