dimitra-maoutsa / neuralflow

The framework for inferring Langevin dynamics from spike data

NeuralFlow

Short description

Computational framework for modeling neural activity with continuous latent Langevin dynamics.

Quick installation: pip install git+https://github.com/engellab/neuralflow

The source code for the following publications:

1. Genkin, M., Hughes, O. and Engel, T.A., 2020. Learning non-stationary Langevin dynamics from stochastic observations of latent trajectories. Nat Commun 12, 5986 (2021).

Link: https://rdcu.be/czqGP

2. Genkin, M., Engel, T.A. Moving beyond generalization to accurate interpretation of flexible models. Nat Mach Intell 2, 674–683 (2020).

Link: https://www.nature.com/articles/s42256-020-00242-6/

Free access: https://rdcu.be/b9cW3

Installation and documentation

https://neuralflow.readthedocs.io/

Tutorial

Part 1: Data format

Convert data from the spike times format to the ISI format.

Part 2: EnergyModel Class

Create EnergyModel class and visualize the framework parameters.

Part 3: Synthetic data generation

Generate synthetic data and latent trajectories from the ramping dynamics and visualize the latent trajectories, firing rate along these trajectories, and the spike rasters.

Part 4: Model Inference

Optimize a model potential on spike data generated from the ramping dynamics.

Part 5: Feature consistency analysis for model selection

Implement feature consistency analysis for model selection.