Combined similarity and QSPR virtual screening for guest molecules of β-cyclodextrin

Abstract

We describe a similarity-based screening approach combined with a quantitative prediction of affinity based on physicochemical descriptors, for the efficient identification of new, high affinity guest molecules of β-cyclodextrin (β-CD). Four known β-CDguest molecules were chosen as query molecules. A subset of the ZINC database with 117 695 molecular entries served as the screening library. For each query the 150 most similar molecules were identified by virtual screening against this library with a graph-based similarity algorithm. Subsequently these molecules were scored by means of a QSPR model. The best-scoring, commercially available molecules were selected for experimental verification (14 in total). Binding free energies were determined by isothermal microcalorimetry (ITC). For three of the four queries, at least one ligand with a higher binding affinity than the corresponding query was found. The approach is a promising high throughput alternative to structure-based virtual screening. While β-CD was chosen as a test case because of its technical relevance and the availability of many binding data, the applied methodology is transferable to other host–guest systems.