RankMHC: Learning to Rank Class-I Peptide-MHC Structural Models.

Area of Science:

• Immunology
• Structural Biology
• Computational Biology

Background:

• Peptide binding to Major Histocompatibility Complex (MHC) class I molecules and subsequent T-cell receptor recognition are vital for immune responses against diseases.
• Accurate identification of peptide antigens is essential for developing therapies for infectious diseases and cancer.
• Peptide-MHC (pMHC) structural modeling is increasingly used, but determining the most representative peptide binding pose from computational ensembles remains challenging.

Purpose of the Study:

• To address the challenge of identifying the correct peptide binding mode within ensembles generated by pMHC structural modeling tools.
• To develop a novel computational method for accurately ranking peptide poses in pMHC structural models.

Main Methods:

• The problem of peptide binding pose identification was framed as a Learning-to-Rank (LTR) task.
• A novel LTR-based predictor, named RankMHC, was developed and trained to predict the most accurate ranking of pMHC conformations.
• RankMHC was evaluated against traditional scoring functions and existing machine learning-based predictors.

Main Results:

• RankMHC demonstrated superior performance in identifying the most accurate peptide binding modes compared to classical peptide-ligand scoring functions.
• The developed predictor outperformed previous Machine Learning (ML)-based binding pose prediction methods.
• RankMHC proved to be versatile, applicable across various pMHC structural modeling tools and protocols.

Conclusions:

• RankMHC offers a significant advancement in accurately predicting peptide binding modes for pMHC complexes.
• This method enhances the reliability of pMHC structural modeling workflows for peptide antigen identification.
• RankMHC provides a valuable tool for advancing the development of peptide-based immunotherapies.