Fragment formal concept analysis accurately classifies compounds with closely related biological activities

Fragment formal concept analysis (FragFCA) for compound classification: Signature fragment combinations for compound classes with closely related biological activity were identified using FragFCA. These combinations are used to accurately classify active test compounds on the basis of fragment mapping. FragFCA can extract class-specific fragment combinations from compounds active against different target families that have signature character and practical utility in compound classification and database searching.Formal concept analysis (FCA), originally developed in information science, has been adapted to identify relationships between fragments of compounds and their biological activity. Here applications of the FragFCA approach with practical utility for medicinal chemistry are explored. Hierarchically derived fragment populations of 24 classes of compounds active against eight target families were subjected to FragFCA analysis, and fragment combinations were identified that distinguished compounds with closely related biological activity from each other. Mapping of signature fragment combinations was carried out to classify active compounds for different target families with high accuracy. The results indicate that compound-class-specific structural information and selectivity determinants are predominantly encoded by fragment combinations, rather than individual fragments. Furthermore, class-specific fragment combinations were successfully applied in similarity searching. The results demonstrate that FragFCA is capable of identifying fragment combinations that differentiate between compound sets with closely related biological activity and that can be used to predict structure-activity relationships.