MHC2AffyPred

MHC2AffyPred predicts binding affinities (BA) of peptides to Major Histocompatibility Complex (MHC) class II molecules using a machine-learning approach that employs structural interaction fingerprints and Moran autocorrelation descriptors to estimate peptide-MHC-II interactions across HLA-DRA1, HLA-DRB1, HLA-DP, and HLA-DQ allotypes.

Key Features:

• Machine Learning Approach: Uses a random forest regressor trained separately for peptide lengths of 9–19 amino acids to predict binding affinities from descriptors derived from site-directed docking into pMHC-II crystal structures.
• Template-Based Conformation Generation: Generates biased peptide conformations by placing peptides into existing pMHC-II complex templates instead of producing large ensembles of conformations.
• Automated Workflow: Implements the computational workflow using Linux shell and Perl scripts to apply models to characterized MHC-II allotypes.

Scientific Applications:

• Immunological research: Estimates peptide-MHC-II binding affinities to inform studies of antigen presentation and immune recognition.
• Vaccine design: Ranks candidate epitopes by predicted MHC-II binding affinity to support vaccine antigen selection.
• Therapeutic development: Aids identification of peptide targets for immunotherapies by predicting MHC-II binding strength across HLA allotypes.
• Viral epitope analysis (SARS-CoV-2): Has been applied to SARS-CoV-2 MHC-II peptides, computing IC50 values with reported correlation coefficients of 0.998 versus NetMHCIIpan v3.2 and 0.570 versus v4.0.

Methodology:

Peptides are docked into crystal structures of pMHC-II complexes via site-directed docking to generate interaction fingerprints and template-based peptide conformations; structural interaction fingerprints and Moran autocorrelation descriptors are calculated for each peptide-MHC-II complex; a random forest regressor is trained for each peptide length category (9–19 aa); workflow is implemented with Linux shell and Perl scripts; performance was evaluated against MHCII3D and NetMHCIIpan with correlation coefficients of 0.612–0.898 for HLA-DRA1/HLA-DRB1 and 0.91–0.98 for HLA-DP/HLA-DQ peptides.