Neural Ensemble Reactivation During Inter-Trial Intervals

Analysis code and data for:

Maggi, S. et al. (2018). Prefrontal cortex ensemble reactivation during inter-trial intervals. bioRxiv preprint

Overview

This repository investigates how neural population activity in the prefrontal cortex (PFC) is reactivated during inter-trial intervals (ITIs) of a rule-switching decision-making task. The analysis spans 50 recording sessions and asks whether ensemble patterns during ITIs carry information about past outcomes and future decisions.

Key analyses include:

• Recall matrices — pairwise similarity of population vectors across ITIs, measuring ensemble reactivation strength
• Residual recall — controlling for temporal autocorrelation via inter-spike-interval (ISI) shuffling
• Multi-classifier decoding — 10 classifiers (KNN, Linear SVM, RBF SVM, Decision Tree, Random Forest, AdaBoost, Naive Bayes, LDA, QDA, Logistic Regression) decode task features from population firing rates
• Prospective and retrospective coding — whether ITI activity predicts upcoming or reflects previous trial outcomes, directions, and cue locations
• Session-type comparisons — learning sessions vs rule-change vs stable performance

Original electrophysiology data from CRCNS pfc-6.

Repository Structure

PFCensemble_ITI/
├── Es_DecoderTrialITI_...py       # Multi-classifier decoder (Python/scikit-learn)
├── Es_Recall_ShuffleISI.m         # Recall matrix computation with ISI shuffle control
│
├── Figure2_DeltaRecall_Reward.m   # Figure 2: Delta recall between error/correct ITIs
├── Figure5_Classifier.m           # Figures 5–6: Classifier accuracy across session types
│
├── F_DeleteCell_spikingCell.m     # Filter neurons by minimum spike count
├── F_SimilarityMatrix.m           # Compute population vector similarity matrices
├── Load_Behavior.m                # Load behavioural data for all 50 sessions
├── Load_CellType.m                # Load neuron classification metadata
├── figure_properties.m            # Shared figure formatting parameters
├── nansem.m                       # Standard error of the mean (NaN-safe)
├── b2r.m                          # Blue-to-red diverging colormap (third-party)
│
├── Data/                          # Behavioural data (50 sessions, date-stamped)
└── Processed Data/                # Pre-computed results (recall matrices, classifier scores)

Quick Start

Reproduce figures (MATLAB)

% Figure 2: ensemble reactivation and reward-driven recall
run('Figure2_DeltaRecall_Reward.m')

% Figures 5–6: classifier accuracy across session types
run('Figure5_Classifier.m')

Run the decoder pipeline (Python)

python Es_DecoderTrialITI_Prospective_Retrospective_FitOriginalDataAndShuffled.py

Methods

Method	Language	Application
Population vector correlation	MATLAB	Recall and similarity matrices
ISI shuffle control	MATLAB	Residual recall (temporal autocorrelation removal)
Leave-one-out cross-validation	Python	Decoder accuracy estimation
10-classifier comparison	Python	Robust decoding of outcome, direction, cue from firing rates
Kolmogorov–Smirnov test	MATLAB	Error vs correct recall distributions
Sign test	MATLAB	Session-level delta recall significance

Dependencies

MATLAB:

• Statistics and Machine Learning Toolbox (kstest2, boxplot)

Python:

• numpy, scipy, matplotlib, seaborn
• scikit-learn (KNeighborsClassifier, SVC, RandomForestClassifier, LinearDiscriminantAnalysis, LogisticRegression, etc.)

Data

The Data/ folder contains behavioural data for 50 recording sessions. The Processed Data/ folder contains pre-computed recall matrices and classifier scores needed to reproduce the figures.

For the complete electrophysiology dataset, see CRCNS pfc-6.

Citation

If you use this code or data, please cite:

Maggi, S. et al. (2018). Prefrontal cortex ensemble reactivation during inter-trial intervals. bioRxiv:187948

License

This project is licensed under the MIT License — see LICENSE for details.