Here we report a new drug design workflow that facilitates the transfer of structure-activity relationships (SARs) and recommends alternative fragments from SAR databases. We first prepare two collections of matched molecular series (MMS) comprising a query set of compounds with their SARs and a set derived from reference SAR databases. The second step detects MMS from the reference SAR sources, which identifies profiles similar to a query MMS according to integrated similarities of scaffold shapes and SAR trends. The third step enumerates new compounds with improved activity profiles compared with a query compound computed using a collaborative filtering algorithm. Our workflow detected direct and latent relationships between a query MMS and those derived from the reference SAR sources. Retrospective application of this workflow to the identification of factor Xa inhibitors yielded recommendations with higher predictive accuracy than a conventional quantitative SAR technique. Moreover, potent S1 binding elements were identified using SAR knowledge independent of information about ligand-protein complexes.Key Words: in silico drug design, de novo drug design, data mining, matched molecular pair
Area of Interest:In silico drug discovery