Cheminformatic and assay-performance profiling of small-molecule screening collections
Clemons, Paul A .
MetadataShow full item record
Quantitative decisions about properties and behavior of compound sets are important in building screening collections for smallmolecule probes and drugs. Decisions about individual compounds typically dominate such discussions: individual compounds pass or fail filtering rules, individual compounds hit or not in assays, etc. In this presentation, we focus on analyses directed at sets of compounds rather than individual members. We start with bioinformatic analysis of natural product and drug targets that motivates the need for new sources of synthetic small molecules. Next, we use sets of molecules from 3 sources (commercial, natural, academic) to show that different computed chemical properties (cheminformatic profiles) provide different chemical intuition about diversity of compound sets, and how quantifying these relationships can provide guidance to synthetic chemists. In the second part, we show that arrays of biological performance measurements (assay-performance profiles) can be used, instead of chemical structure, as a basis for small-molecule similarity, with implications for target identification and lead hopping. To illustrate connections between computed and measured properties, we describe a structured small-molecule profiling experiment in which 15,000 compounds were exposed to 100 different protein-binding assays. We show how different computed molecular complexity and shape descriptors accord with specificity of performance in protein-binding assays. Finally, using the same dataset, we introduce a measure of assay-performance diversity based on information entropy, and show how it might be used to judge relationships between computed properties and performance diversity of compound collections.