SAGERank: Inductive Learning of Protein-Protein Interaction from Antibody-Antigen Recognition using Graph Sample and Aggregate Networks Framework

Numerous experiments and computationally solved antibody-antigen interfaces offer the possibility of training deep-learning models to help predict their biological correlations. Predicting antibody-antigen docking and structure-based design represent significant long-term and therapeutically important challenges in computational biology. We present SAGERank, a general, configurable deep learning framework for antibody design using Graph Sample and Aggregate Networks, which mainly includes ranking docking decoys, detecting binding, and identifying biological interfaces. The model proved its reliability in three different tasks. For both problems ranking docking decoys and identifying biological interfaces, SAGERank is competitive with or outperforms, state-of-the-art methods. Besides, the SAGERank model still showed a high degree of confidence in determining whether the antibody-antigen could bind. All of these demonstrate the versatility of SAGERank for structural biology research. Most importantly, our study demonstrated the real potential of inductive deep learning to overcome small dataset problem in molecular science. The SAGERank models trained for antibody-antigen docking can be used to examine generally protein-protein interaction docking and differentiate crystal packing from biological interface.