Statistical Validation of Compound Structure-Selectivity Relationship Using NCI Cancer Database

Existence of compound structure-activity relationship (SAR) and structure-selectivity relationship (SSR) are the fundamental hypotheses medicinal chemists rely on for lead selection and optimization during drug development. Although SAR has recently been statistically examined, previous studies did not cover SSR, an equally important assumption., For the current study, structural similarities (or distances) between compounds are defined by fingerprint-based Tanimoto distances, while selectivity similarities (or distances) are measured by the correlation between corresponding inhibition profiles available in the U.S. National Cancer Institute (NCI). Under these definitions, the validity of SSR assumption is examined in a statistically rigorous manner. Quantitative SSR similarity thresholds are derived in a tabular form, which in turn can be used to control the false positive rate of the SSR search. As an application of our statistical results, 178 seed compounds with good efficacy and selectivity against 60 cancer cell lines were identified and expanded into a larger set of 5469 potential novel anti-cancer compounds using the principle of SAR and SSR. The final collection of our 5469 compounds that are suspected to have strong inhibition against specific cancers based on our statistical study is web-accessible. This method is superior to random selection based approaches by an expected 2.6 fold enrichment. It can also be applied to discover novel compounds from other sources.