Resolving the Connection between Major Histocompatibilty Complexes and Immune Outcomes using Unsupervised Clustering of Molecular Dynamics Simulations

Abstract

One major bottleneck in the broader application of targeted CD8+ T cell therapeutics is the confident identification of peptides capable of binding to a patient expressed major histone compatibility complex class I (MHC-I) allele. Current, MHC-I peptide binding prediction algorithms consist of neural networks trained on experimentally validated binders for 174 alleles. However, MHC-I molecules are highly polymorphic with more than 10,000 unique annotated sequences exhibiting potentially divergent peptide binding motifs. This translates to a lack in coverage for patients expressing MHC-I alleles not covered by the original training set which is heavily skewed towards non-white populations.To overcome this disparity in coverage, we have sought to develop a framework in which MHC-I alleles likely to exhibit the same peptide binding preferences can be grouped into the smallest number of non-redundant sets or “supertypes”. The ability to group MHC-I alleles by likely binding preference would significantly reduce in the number of individual MHC-I alleles binding motifs that need to experimentally defined in order to provide universal patient coverage. These supertypes were determined through analysis of 500ns simulations of 138 different crystal structures covering 37 MHC-I alleles, and the peptide-MHC binding interface was evaluated using the ROSETTA interface analyzer. Subsequently, we were able to identify and weigh 132 residues within the MHC-I binding pocket based on energy contributions to the MHC-peptide binding. The energy-weighted residues were then combined with BLOSUM80 encoding and the distance of each individual allele was calculated to all other alleles, resulting in a distance matrix that was then used to perform hierarchal clustering. This approach to allele clustering demonstrated the superior ability to group MHC-I alleles associated with disease outcome in Ankylosing spondylitis and HIV1 when compared to sequence alone.