Improved Binding Affinity Prediction Using Non-Covalent Interactions and Graph Integration

Junseok Choe, Keonwoo Kim, Minjae Ju, Sumin Lee, Jaewoo Kang

Abstract: We introduce a novel approach for improving drug-target interaction (DTI) prediction. Our work addresses issues related to model interpretability, protein representation and structural changes in binding complexes in previous drug-target prediction models. We propose utilizing non-covalent residue-residue interactions in protein graphs, formulating an extended form of drug-target link prediction involving non-covalent atom-residue interactions and featuring a graph integration scheme that builds a stronger representation for binding complexes.