PepHLA_Interaction

This project predicts peptide-protein（HLA-A） interactions.
The repository already includes pretrained checkpoints and precomputed features, so you can run inference directly.

You can:

• run demo.py for a single-sample demonstration, or
• run evaluate/test.py for evaluation on the provided dataset.

1. Project Structure

PepHLA_Interaction/
├── demo.py                         # Single-sample inference demo
├── README.md
├── requirements.txt
│
├── ckpts/                          # Pretrained model checkpoints
│   ├── HQNN_AUC_0.8530_AUPR_0.6382.pt
│   ├── CNN_AUC_0.8260_AUPR_0.5608.pt
│   └── CAMP_AUC_0.8653_AUPR_0.6429.pt
│
├── dataset/
│   └── HLA.tsv                     # Main input table for inference/evaluation
│
├── evaluate/
│   └── test.py                     # Batch evaluation script (ACC/AUC/AUPR/confusion matrix)
│
├── models/
│   ├── hqnn.py                     # Default model used by demo/test
│   ├── cnn.py
│   └── camp.py
│
└── data_prepare/
    ├── features_HLA/               # Precomputed feature dictionaries (required for inference)
    │   ├── peptide_seq_feature_dict
    │   ├── peptide_ss_feature_dict
    │   ├── peptide_dis_feature_dict
    │   ├── peptide_phys_feature_dict
    │   ├── protein_seq_feature_dict
    │   ├── protein_ss_feature_dict
    │   ├── protein_dis_pssm_feature_dict
    │   ├── protein_phys_feature_dict
    │   ├── pad_pep_len.npy
    │   └── pad_prot_len.npy
    ├── step1_IUPred2A/             # Disorder prediction intermediate outputs
    ├── step2_PSSM/                 # PSSM intermediate outputs
    └── step3_SSPro/                # Secondary-structure intermediate outputs

Data Mapping Overview

• Each row in dataset/HLA.tsv represents one protein + peptide pair.
• Column order:
  • prot_seq
  • pep_seq
  • label
  • pep_concat_seq
  • prot_concat_seq
• demo.py and evaluate/test.py use these fields as keys to retrieve entries from data_prepare/features_HLA.
• As long as your sample strings exactly match dictionary keys, inference works directly.

2. Environment Setup

Python 3.10+ is recommended.

2.1 Create a Conda Environment (Recommended)

conda create -n pepHLA python=3.10 -y
conda activate pepHLA

2.2 Install Dependencies

pip install -r requirements.txt

Current dependencies:

• torch==2.8.0
• pennylane==0.42.3
• pandas==2.3.3
• numpy==1.26.4
• scikit-learn==1.7.2
• tqdm
• matplotlib

3. Quick Start: demo.py

demo.py is the reader-friendly entry point:

• loads one sample from HLA.tsv,
• fetches all 8 required feature groups from features_HLA,
• loads a checkpoint,
• prints predicted probability and binary class.

3.1 Run with Defaults

python demo.py

Default values:

• checkpoint: ./ckpts/HQNN_AUC_0.8530_AUPR_0.6382.pt
• feature directory: ./data_prepare/features_HLA
• dataset file: ./dataset/HLA.tsv
• sample index: 0

3.2 Select a Different Sample

python demo.py --sample-index 10

3.3 Override Paths

python demo.py \
  --dataset ./dataset/HLA.tsv \
  --feature-dir ./data_prepare/features_HLA \
  --checkpoint ./ckpts/HQNN_AUC_0.8530_AUPR_0.6382.pt

4. Validation Script: evaluate/test.py

evaluate/test.py performs batch evaluation and reports:

• Accuracy / Precision / Recall
• AUC / AUPR
• Optional confusion matrix (printed; optional image output)

Run from project root:

python evaluate/test.py --split test

Options:

• --split {train,test,all} (default: test)
• --cm to print confusion matrix and try saving a figure
• --cm-out to set confusion matrix image path

Example:

python evaluate/test.py --split test --cm --cm-out ./evaluate/confusion_matrix.png

Note: test.py first tries to read dataset/HLA_split.pkl.

5. New Data Processing Reference

This repository already provides ready-to-use HLA features in features_HLA.
For new data, follow the same processing logic and key format.

5.1 Recommended Workflow

1. Prepare a TSV table with the same schema as dataset/HLA.tsv.
2. Generate peptide/protein intermediate features.
3. Convert intermediate outputs into dictionaries with keys matching your TSV fields.
4. Run demo.py first for a single-sample sanity check, then run batch evaluation.

5.2 External Toolchain

• IUPred2A (disorder): https://iupred2a.elte.hu/
• BLAST+ (PSSM): https://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/
• SSPro (secondary structure): https://download.igb.uci.edu/#sspro

5.3 Intermediate File Format Examples

• data_prepare/step1_IUPred2A/*.txt: residue-level disorder scores
• data_prepare/step2_PSSM/prot1.pssm: PSSM matrix example
• data_prepare/step3_SSPro/pep_prot.fasta + pep_prot.ss: sequence/secondary-structure paired examples

---