Building predictive models for mechanism-of-action classification from phenotypic assay data sets

摘要

Compound mechanism-of-action information can be critical for drug development decisions but is often challenging for phenotypic drug discovery programs. One concern is that compounds selected by phenotypic screening will have a previously known but undesirable target mechanism. Here we describe a useful method for assigning mechanism class to compounds and bioactive agents using an 84-feature signature from a panel of primary human cell systems (BioMAP systems). For this approach, a reference data set of well-characterized compounds was used to develop predictive models for 28 mechanism classes using support vector machines. These mechanism classes encompass safety and efficacy-related mechanisms, include both target-specific and pathway-based classes, and cover the most common mechanisms identified in phenotypic screens, such as inhibitors of mitochondrial and microtubule function, histone deacetylase, and cAMP elevators. Here we describe the performance and the application of these predictive models in a decision scheme for triaging phenotypic screening hits using a previously published data set of 309 environmental chemicals tested as part of the Environmental Protection Agency's ToxCast program. By providing quantified membership in specific mechanism classes, this approach is suitable for identification of off-target toxicity mechanisms as well as enabling target deconvolution of phenotypic drug discovery hits.