ASPred: identification of antigen apecific B-cell receptors from single V(D)J sequences using large language models 4171

Abstract Description The rapid sequencing of antibody genes has accelerated vaccine development. However, predicting synthetic antibodies capable of binding and neutralizing novel antigens remains challenging due to a limited understanding of the rules of protein-protein interaction at the surface of an antigen to which its cognate antibody binds. While recent advances in single-cell sequencing of antibody-producing B-cells sequences have improved precision in mapping B-cell receptors to their cognate antigens, there remain additional challenges. We have developed a computational strategy, the Antibody Specificity Predictor (ASPred), with which we have trained two Large Language Models (LLMs) with known sequences of antigen-BCR pairs to predict antigen-specific BCRs from the total BCR repertoire. By leveraging pattern recognition capabilities of LLMs we successfully classify novel B-cell receptors with a challenge antigen not represented in the training set, without the need for preselecting the B cells by antigen binding. The properties of the top 10 predicted candidates were validated by coarse-grained molecular dynamics simulations. These results suggest that sufficient information exists in BCR-antigen sequence pairs for LLMs to reliably predict antigen-antibody interaction specificity, potentially opening new avenues for the computational design of synthetic antibodies for vaccine and therapeutic development. Funding Sources The work described in this report was supported by the National Institute of Allergy and Infectious Diseases (NIAID) of the National Institutes of Health (NIH) under the award number 1R01AI169543 Topic Categories Computational and Systems Immunology (COMP)