A multi-region recurrent circuit for evidence accumulation in rats

Authors: Diksha Gupta, Charles Kopec, Adrian Bondy, Thomas Luo, Verity Elliott, Carlos Brody

Overview

This repository contains analysis code for investigating neural population dynamics in rat frontal orienting fields (FOF) and anterior dorsal striatum (ADS) during a perceptual decision-making task with accumulating sensory evidence. The analyses include neural encoding and decoding, generalized linear models (GLM), reduced-rank regression, optogenetic perturbation experiments, and multi-region recurrent neural network (RNN) modeling.

Repository Structure

├── Code/
│   ├── figure_code/          # Analysis scripts organized by figure
│   │   ├── figure1/           # Population response characterization
│   │   ├── figure2/           # Neural encoding/decoding and GLM
│   │   ├── figure3/           # Optogenetic perturbation analysis
│   │   ├── figure4_6/         # RNN modeling
│   │   └── helpers/           # Shared utility functions
│   ├── saved_results/         # Intermediate analysis outputs (.npy, .pkl)
│   ├── run_params.py          # Global configuration and paths
│   └── my_imports.py          # Standard imports and plotting settings
├── figure_pdfs/               # Generated figures (PDF format)
├── requirements.txt           # Python dependencies

Main Analyses

Figure 1: Population Neural Responses

• Scripts: figure1_popraster.py, figure1_trialraster.py
• Outputs: Population PSTHs, single-trial rasters

Figure 2: Neural Encoding and Decoding

• Scripts: figure2_encodingfigs.py, figure2_decodingfigs.py, figure2/fig2_helpers/
• Analyses:
  • Choice, evidence, and history decoding using logistic regression
  • Tuning curve analysis (compute_hanks_tuning_curves.py)
  • Reduced-rank GLM for inter-region communication (neural_GLM/glmfits_rr.py)
  • Cross-correlation of decision variables (DVcc_sims.py)

Figure 3: Optogenetic Perturbations

• Script: figure3/figure1_metaopto.m (MATLAB)
• Analysis: Effects of inactivation of FOF to ADS projections on behavior

Figures 4-6: RNN Modeling

• Scripts: figure4_6/train_nn_1_multiplicative_opto.py, analysis/fig4_save_*.py
• Architecture: Multi-region RNN with recurrent connectivity trained on evidence accumulation task
• Analyses: Network inactivation simulations, decoding from RNN units, recovery mechanisms

Configuration

Before running analyses, edit Code/run_params.py:

BASEPATH = "/path/to/Code/"

All other paths are automatically configured relative to BASEPATH.

Contact

For questions regarding code or analyses, please contact Diksha Gupta [diksha.gupta@ucl.ac.uk]