• This repository contains the source code for the paper pMHChat, Characterizing the Interactions Between MHC Class II Molecules and Peptides with LLMs and Deep Hypergraph Learning.

pMHChat is developed for MHC II-peptide binding prediction with pLMs and hyperconv.

• BD2016: 5-fold CV dataset(https://services.healthtech.dtu.dk/suppl/immunology/NetMHCIIpan-3.2/)
• BD2024: Processed independent test set(http://tools.iedb.org/auto_bench/mhcii/weekly/)
• BC2015: Crystal Structures of 51 pMHC complex
• checkpoint for fine-tuning stage can be found at https://zenodo.org/records/15057065

• conda create -n py310 python=3.10
• conda activate py310
• pip3 install torch torchvision torchaudio #for cuda 12.4
• conda install pyg -c pyg
• pip install -r requirements.txt

• Driven by binding reactivity prediction task, pretrain and fine-tune ESM-MSA-1B and ESM-2.
• python llm_pretrain/main.py

• Generate the residue-level feature for MHC pseudosequence and peptide sequence, as well as residue contact map of peptide
• create_mhc_features.py
• create_peptide_features.py

• python main.py