ProPepX
ProPepX is a unified deep learning framework for residue-level protein–peptide binding-site prediction. The framework introduces interaction-aware transfer learning with bidirectional cross-attention to jointly model protein and peptide representations, enabling accurate prediction of binding residues on both interaction partners.
ProPepX provides pretrained checkpoints for protein-side, peptide-side, joint, and zero-shot prediction using ProtTransT5 and ESM-3 (600M) protein language models.
Model Variants
| Task | ProtTransT5 | ESM-3 (600M) |
|---|---|---|
| Protein-side prediction | ✓ | ✓ |
| Peptide-side prediction | ✓ | ✓ |
| Joint prediction | ✓ | ✓ |
| Zero-shot prediction | ✓ | ✓ |
Quick Inference
After downloading the corresponding checkpoint and embeddings, run:
python propepx_predict.py \
--protein "<PROTEIN_SEQUENCE>" \
--peptide "<PEPTIDE_SEQUENCE>" \
--embedding {prottrans|esm} \
--mode {prot|pep|mode-GLOBAL|zero-shot} \
--dataset <DATASET_NAME> \
--save_html results/
Example output:
results/
├── prediction_report.html
├── protein_binding_scores.csv
├── peptide_binding_scores.csv
├── interaction_heatmap.png
└── summary.json
Resources
| Resource | Link |
|---|---|
| 🌐 GitHub Repository | https://github.com/SyedKumailHussainNaqvi/ProPepX |
| 🖥️ Interactive Web Server | https://syedkumailhussainnaqvi.github.io/ProPepX/ |
| 📚 Documentation | https://github.com/SyedKumailHussainNaqvi/ProPepX/tree/main/docs |
Intended Use
ProPepX is designed for research applications in:
- Protein–peptide interaction prediction
- Residue-level binding-site prediction
- Structural bioinformatics
- Computational biology
- AI-assisted protein engineering
The models are intended for research use only and should be experimentally validated before downstream biological interpretation.
Citation
If you use ProPepX in your research, please cite:
@article{ProPepX2026,
title={ProPepX: A unified, interpretable, bidirectional interaction-aware transfer-learning framework for residue-level protein–peptide binding-site prediction},
author={Syed Kumail Hussain Naqvi and Sourav Chandra et al.},
journal={Nature Machine Intelligence},
year={2026},
note={Manuscript under review}
}
Contact
Syed Kumail Hussain Naqvi
Department of Physical-AI Convergence Engineering
Jeonbuk National University, Republic of Korea