New Diversity Criterion and Database Compression Method

Based on the topological scaffold classification approach to cluster a structural database, we propose a new criterion to evaluate the diversity of a chemical structural database. This criterion is defined as the ratio of scaffold number to total structure number in the database. Six databases have been evaluated by this criterion. To reduce the size of a database under the minimum losing structural diversity, a novel effective database compression method has been developed. The number of selected structures in each compounds group with common scaffold is determined by empirical K–4–5 rules, and the selected structures are those with the higher drug–like value (DLV). A validity test has been made by adding 200 new random nonoverlapping natural products into NCI3D, and the losing percent of that test data set is about 10.5%. Results show that NCI3D, MNPD (marine natural products database) and TCMD (traditional Chinese medical database) have 68.7%, 60.3%, 54.4% size reduced respectively by this method.