Looking Back to the Future: Predicting <i>in Vivo</i> Efficacy of Small Molecules versus <i>Mycobacterium tuberculosis</i>

Selecting and translating <i>in vitro</i> leads for a disease into molecules with <i>in vivo</i> activity in an animal model of the disease is a challenge that takes considerable time and money. As an example, recent years have seen whole-cell phenotypic screens of millions of compounds yielding over 1500 inhibitors of Mycobacterium tuberculosis (<i>Mtb</i>). These must be prioritized for testing in the mouse <i>in vivo</i> assay for <i>Mtb</i> infection, a validated model utilized to select compounds for further testing. We demonstrate learning from <i>in vivo</i> active and inactive compounds using machine learning classification models (Bayesian, support vector machines, and recursive partitioning) consisting of 773 compounds. The Bayesian model predicted 8 out of 11 additional <i>in vivo</i> actives not included in the model as an external test set. Curation of 70 years of <i>Mtb</i> data can therefore provide statistically robust computational models to focus resources on <i>in vivo</i> active small molecule antituberculars. This highlights a cost-effective predictor for <i>in vivo</i> testing elsewhere in other diseases