Predicting Affinity Through Homology (PATH): Interpretable binding affinity prediction with persistent homology
Fig 4
Given a protein-ligand complex, PATH+ computes internuclear persistence contours (IPCs) using persistent homology, and selects a subset of features into persistence fingerprint, which is then used to predict binding affinity by a sparse set of regression trees (orange). During training (blue), protein-ligand structures with experimentally measured binding affinities from PDBBind are used to derive an optimal set of features for persistence fingerprint and an optimal set of regression trees.