REPLAY Code

Analysis pipeline for detecting sequential neural replay patterns in infant EEG data.

Author: Christopher Postzich License: MIT (SkeideLab 2025)

Overview

This project implements a complete EEG analysis workflow to detect memory replay during rest and learning in infants. The pipeline includes:

• Loading and preprocessing EEG data
• Training logistic regression classifiers on localizer stimuli
• Cross-temporal decoding across experimental segments
• Temporally Delayed Linear Model (TDLM) analysis for sequenceness detection
• Statistical inference via permutation testing
• Publication-quality visualizations

Project Structure

REPLAY_code/
├── load_transform_data.py    # Main analysis pipeline
├── plot_results.py           # Results visualization
├── utils/                    # Utility modules
│   ├── imports.py            # Data import functions
│   ├── utils.py              # Statistical helpers
│   ├── TDLM.py               # Temporal sequenceness analysis
│   ├── CrossDecoding_MEEG.py # Cross-temporal decoding
│   └── plots.py              # Plotting utilities
├── additional_data/          # Data files
│   ├── participant_info.xlsx # Participant metadata
│   └── clusterdepth_pvals.npz# Pre-computed statistics
├── LICENSE
└── pyproject.toml

Installation

Requirements

• Python >= 3.10
• MNE-Python (EEG data handling)
• NumPy, SciPy, Pandas
• Scikit-learn (classifiers)
• Matplotlib, Seaborn (visualization)
• Openpyxl (Excel file reading)

Setup

# Clone the repository
git clone https://github.com/SkeideLab/REPLAY.git
cd REPLAY/Analysis/REPLAY_code

# Install dependencies (using pip)
pip install -e .

# Or install dependencies directly
pip install mne numpy scipy pandas scikit-learn matplotlib seaborn openpyxl

Usage

Main Analysis Pipeline

The primary analysis is performed via load_transform_data.py:

# Run the full analysis pipeline
python load_transform_data.py

This script:

1. Loads preprocessed EEG segments (localizer, resting, cued replay, sequence learning)
2. Trains classifiers on localizer stimuli (Apple, Chair, Face)
3. Applies cross-temporal decoding
4. Computes TDLM sequenceness measures
5. Saves results for visualization

Generating Figures

# Generate publication figures
python plot_results.py

Creates visualizations including:

• Time-resolved decoding curves
• Spatial patterns and topographies
• Sequenceness time-lag plots
• Transition matrix comparisons

Using Individual Modules

from utils.imports import import_preproc_data_replay_things
from utils.TDLM import TDLM, create_transition_matrix
from utils.CrossDecoding_MEEG import CrossDecoding_MEEG

# Load data
participant_info, eeg_data, behavioral_data = import_preproc_data_replay_things(
    segment=['localizer', 'resting'],
    resample_to=100
)

# Create transition matrix for hypothesis testing
trans_matrix = create_transition_matrix(n_states=3, hypothesis='forward')

# Run TDLM analysis
tdlm = TDLM(max_lag=600, bin_lag=8)
sequenceness = tdlm.fit(classifier_probs, trans_matrix)

Experimental Segments

Segment	Description
localizer	Training data with known stimulus labels
resting	Spontaneous activity for replay detection
pre_resting	Baseline before learning
cued_replay	Stimulus-cued reactivation during resting
seq_learn	Activity during sequence presentation

Key Methods

Cross-Temporal Decoding

Trains classifiers at specific timepoints and tests across all timepoints to identify temporal generalization patterns.

TDLM (Temporally Delayed Linear Model)

Detects sequential reactivation by modeling lagged relationships between classifier outputs, identifying forward and backward replay.

Permutation Testing

Cluster-based sign-flip permutation tests for statistical inference on sequenceness measures.

Output

Results are saved to:

• ../Results/ - Processed data arrays (NPZ format)
• ../../Graphics/Results/ - Figures (SVG/PNG format)

Citation

If you use this code, please cite the associated preprint a link.

License

MIT License - see LICENSE for details.

References

Christopher M. Postzich, Johanna Finnemann, Michael A. Skeide bioRxiv 2025.06.12.659246; doi: https://doi.org/10.1101/2025.06.12.659246