MoRE: MHC-Peptide Binding Prediction

Overview

MoRE (MHC-peptide binding pRediction Engine) is a deep learning framework for predicting MHC-peptide binding interactions. This repository contains:

• PPI-activity: MHC class I and class II binding activity prediction
• PPI-site: MHC-peptide binding site prediction (ESM-2 based models)
• pHLA: pHLA binding prediction
• pMHC-TCR: pMHC-TCR binding prediction

Downloading from Hugging Face

All model files, datasets, and code are hosted on Hugging Face:

Repository: https://huggingface.co/MakWaiGong/MoRE

Download all files via CLI

# Install Hugging Face CLI
pip install huggingface_hub

# Download entire repository
hf sync MakWaiGong/MoRE

Or download specific files/folders:

# Download only the chunked model files (43 GB total)
hf download MakWaiGong/MoRE --include "PPI-site/results/_chunks/*"

# Download all code and small models (~13 GB)
hf download MakWaiGong/MoRE \
  --exclude "PPI-site/results/_chunks/**" \
  --exclude "PPI-site/results/more/*.pth" \
  --exclude "PPI-site/results/esm_plain/*.pth" \
  --exclude "PPI-site/results/linear/*.pth"

Git clone (includes LFS for large files)

git clone https://huggingface.co/MakWaiGong/MoRE

Model Downloads

Large Models (>5 GB, split into chunks)

The following models exceed Hugging Face's 5 GB single-file limit and are stored as compressed chunks:

Model File	Original Size	Chunks	Location
PPI-site/results/more/best_pretrain_model.pth	10.79 GB	3	PPI-site/results/_chunks/more__best_pretrain_model.pth.part_aa/ab/ac
PPI-site/results/more/best_model.pth	10.79 GB	3	PPI-site/results/_chunks/more__best_model.pth.part_aa/ab/ac
PPI-site/results/esm_plain/best_esm_plain.pth	10.63 GB	3	PPI-site/results/_chunks/esm_plain__best_esm_plain.pth.part_aa/ab/ac
PPI-site/results/linear/best_model.pth	10.60 GB	3	PPI-site/results/_chunks/linear__best_model.pth.part_aa/ab/ac

How to Download and Reassemble Large Models

After cloning or downloading from Hugging Face:

# Example: reassemble more/best_pretrain_model.pth
cat PPI-site/results/_chunks/more__best_pretrain_model.pth.part_* > PPI-site/results/more/best_pretrain_model.pth

# Example: reassemble more/best_model.pth
cat PPI-site/results/_chunks/more__best_model.pth.part_* > PPI-site/results/more/best_model.pth

# Example: reassemble esm_plain/best_esm_plain.pth
cat PPI-site/results/_chunks/esm_plain__best_esm_plain.pth.part_* > PPI-site/results/esm_plain/best_esm_plain.pth

# Example: reassemble linear/best_model.pth
cat PPI-site/results/_chunks/linear__best_model.pth.part_* > PPI-site/results/linear/best_model.pth

Or use the provided script scripts/merge_chunks.sh:

bash scripts/merge_chunks.sh

Repository Structure

MoRE/
├── PPI-activity/        # MHC-peptide binding activity prediction
│   ├── data/           # Training and test datasets (CSV)
│   ├── src/            # Source code
│   └── result/         # Trained models (131 MB each)
├── PPI-site/           # MHC-peptide binding site prediction
│   ├── data/           # Datasets
│   ├── src/            # Source code
│   ├── results/         # Trained models
│   │   ├── more/       # Large models (10.79 GB each, chunked)
│   │   ├── esm_plain/ # ESM-2 based models (10.63 GB, chunked)
│   │   ├── linear/    # Linear models (10.60 GB, chunked)
│   │   └── _chunks/  # Chunked model parts for download
│   └── scripts/        # Utilities
├── pHLA/              # pHLA binding prediction
│   ├── data/
│   ├── src/
│   └── result/         # Trained models (131 MB each)
├── pMHC-TCR/          # pMHC-TCR binding prediction
│   ├── data/           # Including tcr.csv (728 MB)
│   ├── src/
│   └── result/         # Trained models (132 MB each)
└── scripts/            # Shared utilities

Citation

If you use MoRE in your research, please cite:

@misc{MoRE2026,
  author = {MakWaiGong},
  title = {MoRE: MHC-Peptide Binding Prediction Engine},
  year = {2026},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/MakWaiGong/MoRE}}
}