Calcium imaging latent variable analysis (CILVA)

Python code for fitting the calcium imaging latent variable analysis model. See the companion preprint at https://www.biorxiv.org/content/10.1101/691261v1 for more information.

Overview

This code fits a latent variable model to a matrix of fluorescence traces. The fitted model provides

• a description of low dimensional latent structure underlying spontaneous activity,
• a decomposition of the fluorescence traces into their evoked and (low dimensional) spontaneous components,
• estimated stimulus filters for each neuron that are unbiased by the modelled spontaneous activity.

Usage

Basic model fitting

Fit the model by calling run.py with the required arguments: python cilva/run.py --data $data --L $L --num_iters $num_iters --iters_per_altern $iters_per_altern --max_threads $max_threads --out $out --tau_r $tau_r --tau_d $tau_d --imrate $imrate where

• $data is the path to the training data. The code expects an NxT matrix of fluorescence levels $data.ca2 and a KxT matrix of stimulus onset times $data.stim. These can simply be renamed text files.
• $L is the number of latent factors.
• $num_iters is the number of times we alternate between estimating the latent variables and the model parameters.
• $iters_per_altern is the number of optimiser steps for calculating the MAP estimate of the latent variables.
• $max_threads configures the number of threads available for multithreaded processing.
• $out is the prefix of the output folder containing the fitted model parameters.
• $tau_r and $tau_d are the rise and decay time constants for the calcium transients.
• $imrate is the imaging rate (or sampling frequency) of the fluorescence microscope. This is used to estmate the imaging noise variance. More several additional arguments are provided in the run.py file.

Example

The Jupyter notebook example.ipynb provides an example of how to fit the model to data and decouple the fluorescence traces (recommended). It can be viewed here:

https://nbviewer.jupyter.org/github/GoodhillLab/CILVA/blob/master/example.ipynb

Cross validation

The code also supports a form of single-fold cross validation: the model parameters and latent variables are learned from the training data specified with the --data argument, we then reinfer the latent variables on held-out test data using the already-estimated model parameters. Test data is supplied with the --test argument.

Output

When CILVA has finished running, the estimated parameters are written to the directory

./$out_L_$L_num_iters_${num_iters}_iters_per_altern_${iters_per_altern}_gamma_$gamma_tau_r_${tau_r}_tau_d_${tau_d}_imrate_${imrate}

where each $arg corresponds to an argument from the Basic model fitting section above.

Requirements

This code requires Python 3.5+ and the NumPy (tested on v1.16.2) and SciPy (tested on v1.1.0) packages.

Contact

Feedback or questions about the code can be directed to marcus.triplett@columbia.edu.