Task-specific topographical maps of neural activity in the primate lateral prefrontal cortex

This repository contains code in support of the paper "Task-specific topographical maps of neural activity in the primate lateral prefrontal cortex"

Publications

Preprint: "Task-specific topographical maps of neural activity in the primate lateral prefrontal cortex

Prerequisites

MATLAB 2023a with the following libraries:

• Curve Fitting Toolbox version 3.9
• Image Processing Toolbox version 11.7
• Statistics and Machine Learning Toolbox version 12.5

Python 3.10 with the following libraries

• torchvision version 0.15.2
• torch version 2.0.1
• numpy version 1.24.2
• matplotlib version 3.7.0

The analysis code was tested on MacOS 10.15.7 with 32G RAM. The expected run time of the pipeline using simulated data on a normal desktop computer should be under 10 minutes.

Data

Simulated data can be found in results/spikeTuningVectors/, which can also be generated by running the script analysis/START_R1_simulate_data.m.

The reconstructed cortical patch from existing histology work can be found in results/goldman_rakic/

Code

Simulating data

START_R1_simulate_data.m This function simulates data for the analysis pipeline. Data will be drawn from a normal distribution with some degrees of spatial autocorrelation imposed. In the simulation, there are 3 tasks, 2 subjects, each having 2 arrays. Three measurement sessions will be generated for each subject in each task.

Visualizing tuning similarity on the array

START_B2_spikesMDS.m This function visualizes tuning similarity of channels on the array using multi-dimensional scaling.

Example usage: START_B2_spikesMDS('taskStr', 'KM')

Estimating the spatial autocorrelation function of tuning profiles

START_B3_donutACF.m and START_B3_donutACF_fitting_summary.m. The first function estimates the spatial autocorrelation of tuning profiles for each array in each session in each task. The second function fits a Laplacian to the spatial autocorrelation function.

Example usage: START_B3_donutACF('taskStr', 'KM') then START_B3_donutACF_fitting_summary('taskStr', 'KM')

Smoothing the array maps with a 2D kernel whose FWHM matches the fitted Laplacian

START_B4_figure_procrustes.m This function smoothes the array maps with 2D kernels whose FWHM matches the fitted Laplacian.

Testing for the consistency of functional topography across time within a task and across tasks

START_B5_daysAcrossTasks.m computes the interval between measurement sessions. START_B6_topoInference_generation.m splits the signal into task related and residuals and computes the correlation matrices (functional topography) for each component for each array in each session in each task. START_B6_topoInference_t_test.m computes the consistency of functional topography and performs statistical inference.

Worth mentioning, in the simple simulation script START_R1_simulate_data.m, the signal is generated at random, without modeling task effects. As a consequence, the results for the inference performed on simulated data are likely to be different from actual findings in the manuscript.

Symbols in the inference results figure:

• light blue dots indicate that the consistency of topography for task related signal is significantly different from that for residuals.
• magenta dots indicate that the consistency of topography is higher than chance under channel permutation tests.
• horizontal lines and error bars in cyan show the mean and standard error of the permutation distribution.
• blue horizontal lines show the noise ceiling for between-task comparisons.
• red dots show that the between-task consistency is significantly lower than noise ceiling.

Random sampling from existing histology work

results/goldman_rakic/START_A1_generate_dataset.py generates random samples of 4 mm x 4 mm structural maps from existing histology work. analysis/START_B7_Goldman_Rakic_donutACF.m down-samples the structural maps to 10 x 10, and estimates the spatial autocorrelation function for these maps. analysis/START_B8_all_donutACF_curves.m summarizes all the spatial autocorrelation functions for both simulated data and sampled structural maps.