translocator-models

Introduction

This repository contains scripts and modules used for modeling and analyzing fold changes in population firing rates in the visual cortex under various stimulus conditions. The scripts implement rate-based models, fitting procedures, and data visualization tools used in Velez-Fort 2025. If you have any questions, please open an issue.

Usage

Dependencies are declared in the pyproject.toml file. To create a virtual environment and install the dependencies, run the following commands:

conda create -n translocator-models python=3.11
conda activate translocator-models
pip install .

To run the script that fits the model to the data and visualizes the results, use the following command:

python translcator-models/fit_the_model.py

Data Description

Datasets

visual_flow: Mean fold changes for visual flow stimuli with 39 clusters.
passive_same_luminance: Mean fold changes for passive same luminance stimuli with 151 clusters.
matched_dataset: Re-balanced fold changes in a matched dataset, including conditions simulating slip.

Conditions

V: Visual flow only.
VT: Visual flow with passive translation.
RV: Visual flow with static running.
RVT: Locomotion with visual flow.
T: Passive translation with visual static stimulus.
RV_slip: Running with increased visual flow (simulating slip).
RVT_slip: Locomotion with visual flow and translation slip.

Model Description

Rate-Based Model

The rate-based model computes the fold change in population firing rate based on stimuli for visual flow (VF), translation (T), and running (R). It uses adaptive responses and combines them with weights and thresholds to model the neural population response.

The model equations:

$$v = w_1(VF > 0) + w_2 ((T-R) > 0)(T-R) + w_3R$$ $$\hat fc(v) = \alpha v + c$$

where:

$v$ is the neuron rate response above baseline in arbitrary units,
$w_1$, $w_2$, and $w_3$ are weights applied to each component of the response,
$\alpha$ is a scaling factor, and $c$ is a constant offset.

In this formulation:

The expressions $(VF > 0)$ and $((T - R) > 0)$ act as adaptive thresholds, activating their respective terms only if the stimulus exceeds a certain threshold.
The resulting value, $v$, represents the weighted, thresholded combination of these stimuli, and the final fold change $\hat fc(v)$ scales and offsets this response.

In the article we consider the following weights:

$w_1 = 1$
$w_2 = 0.6$
$w_3 = 1$
$\alpha = 0.8$
$c = 1$ which simplifies the model equations to:

$$v = (VF > 0) + 0.6((T-R) > 0)(T-R) + R$$ $$\hat fc(v) = 0.8v + 1$$

These weights can also be optimized to fit the model to the data, leading to a similar result.

Arithmetic Sum Model

The arithmetic sum model predicts the fold change by combining empirical fold changes using the formula:

$$\hat fc(VF + T) = \beta_0 + \beta_1 fc(T_{VS}) + \beta_2 fc(VF)$$

$fc(T_{VS})$ is the fold change from translation with constant luminance.
$fc(VF)$ is the fold change from visual flow.
$\beta_0$, $\beta_1$, and $\beta_2$ are weights and an intercept for the model that have been previously fitted to the data.

Functions Overview

Model Functions (models.py):
- adaptive_response: Processes stimuli based on thresholds.
- rate_based_model: Computes fold changes using the rate-based model.
- arithmetic_sum_model: Predicts fold changes using the arithmetic sum model.
Fitting Functions (fitting_methods.py):
- fit_fold_changes_to_data: Fits the rate-based model to experimental data.
- get_predicted_fold_changes_*: Compute predicted fold changes for different datasets.
Main Script Functions (fit_the_model.py):
- fit_and_print_results: Fits the model to data, prints results, and computes arithmetic sum predictions if applicable.

neuroinformatics-unit/translocator-models