MotifLeadDB: A Hierarchical Structural Data Set for Congeneric Ligand Binding Activity Change

Predicting ligand binding free energy change by chemical substitutions is a critical task in drug optimization studies. Despite many efforts, existing computational methods show limited prediction accuracy, with one of the major reasons originating from poor training data. As a first step toward overcoming this challenge, we constructed MotifLeadDB, a structural model data set specially designed for the hit-to-lead optimization scenarios. For each receptor, known congeneric binder ligands were grouped by common scaffolds, and their complex structures were built with both ligand and side-chain optimization, so that the structural basis for the energy change upon functional group substitution could be directly inferred. To enhance the model confidence, ligand docking was guided by experimental complex structures sharing chemical similarity. The database consists of 342,489 protein-ligand complex structural models, which are grouped into 27,077 ligand scaffolds and are bound to 357 nonredundant receptors. The resulting structural models are also provided along with their confidence levels. We expect that the data set can be readily used for case studies to understand structure-activity relationships for target receptors of interest, as well as for training a computational model for estimating structure-based binding activity change.