PreyPursuit

Overview

code for running modeling for: Complementary roles for hippocampus, anterior cingulate and orbitofrontal cortex in composing continuous choice

Contents

• Data availability
• Authorship Statement
• Package Installation
• Behavioral Modeling
• Model Recovery Analysis
• Switch Analysis
• Neural GAM modeling
• dPCA
• Clustering

Data Availability

This repository includes example behavioral and neural data from one participant, provided to demonstrate the full analysis pipeline. It also includes synthetic datasets with matched structure to support population level neural analysis demos.

Authorship Statement

All behavioral modeling code was originally written by Justin Fine, with subsequent modifications and extensions by Assia Chericoni and Taha Ismail. The dPCA implementation is based on the method introduced by Brendel et al., with task-specific adaptations.

System Requirements

Hardware

No special hardware requirements beyond a standard workstation.

Software

The code can be run on major operating systems (macOS, Linux, Windows) after installing the required dependencies.

• Python (3.11), see environment.yml for a complete list of required packages
• MATLAB (for plotting)

Package Installation

Install the required Python packages using the provided conda environment file (will have to install pip packages manually): NOTE: On Windows, Visual Studio C++ Build Tools must be v14.0.0 or greater

conda env create -f environment.yml
conda activate pac_control_env
pip install --no-build-isolation "ssm @ git+https://github.com/lindermanlab/ssm"

Repo Contents

Behavioral Modeling

To run an example of the controller modeling pipeline:

python -m scripts.run_controller_modeling -c pv 

pv example controller class. you can specify any controller from the following list:

• p
• pv
• pf
• pvi
• pif
• pvf

NOTE: all results will by default be stored in the example_data/ directory named by controller output (e.g., example_data/YFD/pv). Folder will contain model_fit_results.csv (model weights + ELBO) and model_matrices.mat (w_t timeseries by trial)

The controller modeling pipeline has a runtime of approximately 15 minutes and outputs trial-wise predicted w_t trajectories for each specified controller class.

Model Recovery Analysis

Code for simulation-based model recovery analysis is provided in demo/simulation_demo.ipynb. Due to long runtimes, example results and MATLAB plotting scripts are included in plotting/simulation.

Switch Analysis

Switch detection is performed using the functions in the ChangeOfMind/functions/processing.py script. An example implementation is provided in the demo/ folder as a Jupyter notebook (demo/plot_switch_detection.ipynb). This demo can be run directly on the output of the behavioral modeling pipeline (i.e., trial-wise w_t time series). The GLM used for the switch statistics can be found under legacy/MAIN_GLM_all_subj.ipynb.

Neural GAM modeling

To run GAM modeling with example data, execute the following script

python -m scripts.run_glm_modeling -p YFD

^ can optionally specify controller type to use for w_t with -m <c>

can plot proportion of tuned neurons, NNMF results, and fischer information results using the plotting notebook included under plotting/plot_nmf_fi.ipynb

dPCA

Neural population analysis include condition-based and w_t-binned dPCA. Full dPCA analysis can be found under src/dPCA and run with the DPCAPipeline. The example notebooks in demo/dPCA_demo_Final.ipynb use synthetic data to demonstrate the full pipeline. Plotting scripts can be found in plotting/MAIN_plot_dPCA_results.m

Clustering analysis

The LDS + KMeans clustering pipeline is implemented in scripts/lds_cluster_pipeline.py. The example notebook demo/run_lds_cluster_pipeline_demo.ipynb demonstrates the analysis on data from a single subject for clarity, although the full pipeline is designed to operate on pseudopopulations pooled across subjects.