Fig. 1
From: Prediction of antibody-antigen interaction based on backbone aware with invariant point attention
Schematic overview of AbAgIPA. (A) Problem motivation. Antibody-antigen binding is directly determined by paratope conformation and candidate epitope conformation. (B) Representation construction. Inputs are antibody sequences and candidate epitope structures, establish backbone framework for the antibody structure obtained by IgFold and the known antigen structure, use amino acid physicochemical properties as residue node features. (C) Based on the residue node feature and backbone frame, Input the amino acid node features and backbone frame of antibody and antigen, and utilize invariant point attention layer and global max pooling to get the interaction pattern feature vector of antibody and antigen, in which CDR region mask is used in extracting the antibody interaction pattern feature vector. Further, the interaction labels were obtained using feature splicing and fully connected layer