Data Mining Meets Machine Learning: A Novel ANN‐based Multi‐body Interaction Docking Scoring Function (MBI‐score) Based on Utilizing Frequent Geometric and Chemical Patterns of Interfacial Atoms in Native Protein‐ligand Complexes

Abstract Accurate prediction of binding poses is crucial to structure‐based drug design. We employ two powerful artificial intelligence (AI) approaches, data‐mining and machine‐learning, to design artificial neural network (ANN) based pose‐scoring function. It is a simple machine‐learning‐based statistical function that employs frequent geometric and chemical patterns of interacting atoms at protein‐ligand interfaces. The patterns are derived by mining interfaces of “native” protein‐ligand complexes. Each interface is represented by a graph where nodes are atoms and edges connect protein‐ligand interfacial atoms located within certain cutoff distance of each other. Applying frequent subgraph mining to these interfaces provides “native” frequent patterns of interacting atoms. Subsequently, given a pose for a protein‐ligand complex of interest, the pose‐scoring function (the information‐processing unit or neuron) calculates the degree of matching between the interaction patterns present at the pose's interface and the native frequent patterns. The pose‐scoring function takes into account the frequency of occurrence of the matching native patterns, the size of the match, and the degree of geometrical similarity between pose‐specific and matching native frequent patterns. This novel “multi‐body interaction” pose‐scoring function (MBI‐Score) was validated using two databases, PDBbind and Astex‐85, and it outperformed seven commonly used commercial scoring functions. MBI‐Score is available at www.khashanlab.org/mbi‐score .