A data-driven journey using results from target-based drug discovery for target deconvolution in phenotypic screening

In drug discovery, various approaches exist to find compounds that alter the different states in living organisms. There are two fundamental discovery strategies regarding the mechanism of action: target-based and phenotypic methods. Both have strengths and weaknesses in assay development, target selection, target validation and structure optimization. While phenotypic screening can identify chemical starting points with the desired phenotype, it is typically difficult to carry out efficient, structure-based optimization without confirming the mechanism of action of such hits. It is therefore critical to uncover the targets behind the phenotype. Target deconvolution is typically carried out by a set of highly selective compounds, where each ligand is associated with a particular target. Hits of such a high-selectivity set can provide valuable information on the phenotype's underlying targets and may also enable novel target-based therapeutic strategies. Consequently, there is a continuously high demand for novel highly-selective tool compounds for target deconvolution. In this work, the ChEMBL database, comprising over 20 million bioactivity data, was mined to identify the most selective novel ligands for a diverse set of targets. A novel method for the automated selection of such high-selectivity ligands is presented. Using these high-selectivity compounds in phenotypic screening campaigns can provide a valuable preliminary direction during target deconvolution. 87 representative compounds were purchased and screened against 60 cancer cell lines. Several compounds were found to possess selective inhibition of cell growth of a few distinct cell lines. The phenotypic assay results, along with the nanomolar activities of individual proteins obtained from the ChEMBL database suggest some novel mechanisms of action for anti-cancer drug discovery.