The invention provides a complementary determining region 3 and immune
epitope binding prediction method, and belongs to the field of
bioinformatics. According to the method, multi-
modal heterogeneous graph modeling, a graph
attention network (GAT) and multi-objective
loss function collaborative optimization are fused, and the method aims at breaking through the limitation of the prior art in the aspects of heterogeneous
data modeling,
class imbalance, prediction precision and the like. A heterogeneous graph of 3 nodes of a complementary determining region and immune
epitope nodes is constructed, GAT is introduced to realize cross-
modal feature interaction, and a dynamic weight optimization strategy of a focus
loss function and an AUC
loss function is combined, so that the recognition capability and the overall sorting performance of the model on difficult samples are improved. And meanwhile, interpretable analysis combined with the hotspot residues is realized through the attention weight. According to the method, the prediction accuracy of the binding specificity of the
complementarity determining region 3 and the immune
epitope is remarkably improved, and the method can be widely applied to
cancer vaccine design, individualized
immunotherapy and
autoimmune disease research and has important theoretical significance and practical application value.