Mapping of Activity-Specific Fragment Pathways Isolated from Random Fragment Populations Reveals the Formation of Coherent Molecular Cores

Abstract

Randomly generated molecular fragment populations are investigated as a source for compound-class-dependent chemical information. The analysis of conditional probabilities of fragment co-occurrence in random fragment populations reveals that different classes of active compounds produce series of fragments whose presence depends on each other. Such relationships constitute a fragment hierarchy that becomes a signature of a compound class. We find that such sets of fragments isolated from random populations are typically found to form coherent molecular cores in active compounds. Thus, class-specific random fragment hierarchies encode meaningful structural information. Characteristic core regions already formed by small numbers of substructures remain stable when more fragments are added. These findings provide a structural rationale for the signature character of activity-specific fragment pathways. Thus, randomly generated fragment populations can be mined for combinations of substructures that characterize activity classes. It follows that compound-class-directed structural descriptors can be isolated from random fragment populations that do not depend on the application of predefined fragmentation or design schemes.