A deep-learning framework for designing the complementarity-determining regions (CDRs) of an antibody that jointly generates their amino-acid sequence and 3D structure. It represents the antigen-antibody complex as an attributed heterogeneous graph with E(3)-invariant node and edge features and eight edge relations, then runs relation-aware equivariant message passing that treats antigen-antibody interaction as a dynamic edge-relation optimization, so it needs no pre-specified epitope. It further adds a contrastive specificity-enhancing objective and a metric scoring binding to target versus non-target antigens, addressing the tendency of prior optimizers to produce non-specific antibodies.
Antibodies are Y-shaped proteins that protect the host by binding to specific antigens, and their binding is mainly determined by the Complementary Determining Regions (CDRs) in the antibody. Despite…