Model Building of Antibody–Antigen Complex Structures Using GBSA Scores

Structure prediction of antibody–antigen complexes, which involves molecular docking to generate decoys that are ranked using a scoring function, is an important approach in the design of antibody drugs and biosensors. However, it is not easy to evaluate the stability of protein–protein complexes, using a single scoring function. Here, we developed a prediction method of antibody–antigen complex structures using the docking engine "surFit" and a scoring function (GBSA score) that combined the generalized Born (GB) energy and the hydration energy based on the solvent-accessible surface area (SA). We chose 95 antibody–antigen structural datasets for self-docking and generated many decoy structures using the surFit program. The GBSA scores were computed for all of the decoys, and the area under the curve (AUC) of the GBSA scores yielded a higher value (0.972) than the values obtained by the original surFit scores (0.873) and the ZRANK scores (0.953). To improve the accuracy of prediction, molecular dynamics (MD) simulations were performed for several decoy structures that had good GBSA scores. Consequently, average GBSA scores from MD trajectories can reduce the number of non-native decoys that have GBSA scores competitive with the near-native ones.