Computational design of class II MHC binding peptide with sequence-based evolution information

Abstract

MHCII-peptide binding plays a vital role in immunology. MHCII molecules are primarily expressed on the surface of antigen presenting cells where they capture and present exogenous antigen peptides to helper T cells, thereby activating humoral and cellular immune responses. Designing artificial MHCII binding peptides mimicking native peptides is crucial for vaccine development. However, it is challenge to design artificial binding peptides with conventional structure-based methods since the structures of the peptides are highly flexible at unbound state. In this work, we trained Transformer neural network to design the artificial peptides with sequence-based evolutionary information including the frequency distribution of amino acids at each site and the joint frequency distribution between amino acid pairs extracted from multiple sequence alignment of native peptides. In light of accurate sequence-based scoring function and reliable AlphaFold3 for complex structure prediction, the designed artificial peptides were predicted to have comparable binding affinities as native peptides and high structural confidence (pLDDT > 90.0) when binding to MHCII. Our work establishes a paradigm for designing different kinds of functional peptides and will greatly provide significant assistance to biomedical researchers in the medical and industrial fields.