Data Science & Data Skills for Neuroscientists (J.W. Pillow, SfN 2016)

The material proposed in these tutorials revisits what was presented by Jonathan Pillow in a “short course” on Data Science and Data Skills for Neuroscientists organized at the SFN 2016 meeting, constructing and fitting models with pynapple and NeMoS.

The original Matlab implementation and its python translation can be found at the following links:

• Matlab: pillowlab/GLMspiketraintutorial

• Python: pillowlab/GLMspiketraintutorial_python

Contents

• Tutorial 1 - Poisson GLM
  • Downloading the dataset
  • Loading data into pynapple
  • Pre-processing
  • Building the design matrix
  • Compute and visualize the spike-triggered average (STA)
  • Whitened STA
  • Rate prediction with a linear-Gaussian GLM
  • Linear-Gaussian GLM with NeMoS
  • Poisson GLM
  • Non-parametric estimate of the nonlinearity
  • Quantifying performance: log-likelihood
  • Quantifying performance: AIC
  • Simulating from the GLM

• Tutorial 2 - Poisson GLM with Spike History and Coupling
  • Load and pre-process the data
  • Compute and plot the cross-correlogram
  • Building the design matrix
  • Fit a single-neuron GLM with spike-history
  • Fit coupled GLM for multiple-neuron responses
  • Comparing the models

• Tutorial 3+4 - Gaussian and Poisson GLM with Regularization
  • Load and pre-process the data
  • Upsample to get finer timescale representation of stim and spikes
  • Fit the linear-Gaussian model using ML
  • Ridge regression
  • NeMoS advanced: L2-smoothing
  • Regularizers shipped with NeMoS

• Tutorial 5 - MAP Decoding
  • Introduction
  • The MAP decoding framework
  • Load and pre-process the data
  • Train / test split
  • Fit the forward models
  • Define the smoothness prior
  • MAP objective functions
  • Optimize and decode
  • Evaluate the decoded stimulus

What’s changed

• Tutorial 3 and 4 on regularization have been merged into a single one.

• In the regularization tutorial, an example is shown on how to smooth over multiple predictors.

Citation

If you found this material useful and you wish to cite this tutorial, feel free to :

• Acknowledge the paper from which it was developed: Pillow et al, Nature 2008

• Acknowlege pynapple package by citing the accompanying paper.

• Acknowledge NeMoS package by citing the associated DOI.