A Subgroup Discovery Approach for Relating Chemical Structure and Phenotype Data in Chemical Genomics

We report on development of an algorithm that can infer relations between the chemical structure and biochemical pathways from mutant-based growth fitness characterizations of small molecules. Identification of such relations is very important in drug discovery and development from the perspective of argument-based selection of candidate molecules in target-specific screenings, and early exclusion of substances with highly probable undesired side-effects. The algorithm uses a combination of unsupervised and supervised machine learning techniques, and besides experimental fitness data uses knowledge on gene subgroups (pathways), structural descriptions of chemicals, and MeSH term-based chemical and pharmacological annotations. We demonstrate the utility of the proposed approach c 2010 Umek et al.. in the analysis of a genome-wide S. cerevisiae chemogenomics assay by Hillenmeyer et al.