Greater than the sum of its parts: combining models for useful ADMET prediction

Abstract

In silico ADMET (absorption, distribution, metabolism, excretion, and toxicity) models are important tools in combating late-stage attrition in the drug discovery process. This work shows how ADMET models can be combined to tailor predictions depending on one's needs. We demonstrate how the judicious use of data and considered combination of predictions can produce models that provide truly useful answers. This approach is illustrated with the prediction of hERG channel blocking and cytochrome P450 2D6 inhibition, where combination of two predictive models (with >80% of compounds correctly predicted) resulted in models with even better predictive values (with >90% of compounds correctly predicted for those classes of interest).