Related Experiment Video
Updated: Jun 16, 2026
06:50
Computational Prediction of Amino Acid Preferences of Potentially Multispecific Peptide-Binding Domains Involved in Protein-Protein Interactions
Published on: January 26, 2024
PMGen: from peptide-MHC structure prediction to peptide generation
Amir H Asgary1, Amirreza Aleyasin1, Jonas A Mehl1
1Quantitative and Computational Biology Group, Max Planck Institute for Multidisciplinary Sciences, Göttingen 37077, Germany.
Bioinformatics (Oxford, England)
|June 15, 2026
Summary
PMGen accurately predicts peptide-MHC structures and designs novel peptides for immunotherapy. This framework improves upon existing methods for peptide-MHC modeling and design.
Area of Science:
- Computational biology
- Structural biology
- Immunoinformatics
Background:
- Accurate peptide-MHC (pMHC) structural modeling is crucial for designing immunotherapies.
- Current prediction tools have limitations in coverage, peptide length, and accuracy.
- Existing design strategies often neglect spatial and biophysical insights from pMHC structures.
Purpose of the Study:
- To introduce PMGen, an integrated framework for predicting and designing variable-length peptides in peptide-MHC complexes.
- To enable structure-guided design of peptides for enhanced immunotherapy applications.
- To overcome limitations of existing pMHC modeling tools.
Main Methods:
- PMGen utilizes AlphaFold2 with enforced anchor constraints via Initial Guess and Template Engineering strategies.
- The framework supports variable-length peptides across MHC class I and II.
- ProteinMPNN sampling is employed on predicted backbones for peptide design.
Main Results:
- PMGen achieves state-of-the-art structural fidelity without model fine-tuning.
- Outperforms existing methods with median peptide-core Cα RMSDs of 0.62 Å (MHC-I) and 0.33 Å (MHC-II).
- Successfully recovers incorrect anchor positions and captures mutation-induced conformational changes.
- ProteinMPNN sampling on PMGen structures yields higher-affinity peptides.
- Significantly improves ProteinMPNN's peptide sequence recovery for unseen MHC-I alleles.
Conclusions:
- PMGen provides a powerful tool for accurate pMHC structure prediction and structure-guided peptide design.
- The framework enhances immunotherapy design by leveraging spatial and biophysical insights.
- Accurate predicted structures from PMGen are valuable for downstream machine learning tasks.
