A Systematic Computational Method to Predict and Enhance Antibody-Antigen Binding in the Absence of Antibody Crystal Structures

Abstract

Structure-based design efforts on antibodies are frequently hamstrung by the general difficulty of crystallizing antibody molecules. Computational approaches are getting more success and playing a very important role in protein structure prediction and design. A combined platform with both computational predication and experimental verification would have a transformative research capability on antibody engineering. We have been developing and integrating RosettaAntibody (homology modeling and H3 loop modeling), RosettaDock (SnugDock and EnsembleDock) and RosettaDesign (non-canonical amino acids design) into a systematic computational pipeline. Using the targets of M18/Anthrax complex, which has solved crystal structure, and anti-MS2/MS2 complex, whose antibody structure is not available, the new pipeline is capable of building homology models for antibody from its sequence, and predicting the docked antibody-antigen structures. The correct antigen epitope can be identified by iteratively refining the predicted candidates with experimental verifications such as interface alanine scanning. This is followed by computational CDR designs utilizing non-canonical amino acids, which can form covalent bonds with epitope residues. Along with experimental confirmations, the incorporation of the systematical computational method was proved to significantly facilitate the epitope identification and non-canonical amino acid designs. These efforts eventually substantially increased the antibody-antigen binding affinity of the two targets.