Efficient peptide-MHC-I binding prediction for alleles with few known binders

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning

Background:

• In silico prediction of peptide-MHC-I binding is crucial for T-cell epitope discovery.
• Existing statistical and machine learning methods struggle with polymorphic MHC alleles lacking sufficient binding data.
• Leveraging cross-allele information can enhance prediction performance.

Purpose of the Study:

• To develop a novel support vector machine (SVM) algorithm for simultaneous peptide-MHC-I binding prediction across multiple alleles.
• To improve the accuracy of T-cell epitope identification by sharing binding information across related MHC alleles.

Main Methods:

• Designed a support vector machine algorithm that learns peptide-MHC-I binding models by sharing information across alleles.
• Information sharing is controlled by a user-defined similarity measure between alleles, based on supertypes or key residue comparisons.
• Evaluated the method on benchmark experiments against state-of-the-art approaches.

Main Results:

• The proposed method demonstrates superior performance compared to existing state-of-the-art methods in benchmark experiments.
• Successfully learned peptide-MHC-I binding models for multiple alleles simultaneously by effectively sharing binding data.
• The allele similarity measure, based on supertypes or key residues, proved effective in guiding information sharing.

Conclusions:

• The developed cross-allele learning approach significantly improves the prediction of peptide-MHC-I binding.
• This method offers a robust solution for identifying T-cell epitopes, especially for MHC alleles with limited data.
• The approach is implemented as a publicly available web server and associated code.