Predicting Molecular Targets for Small-Molecule Drugs with a Ligand-Based Interaction Fingerprint Approach

The computational prediction of molecular targets for small-molecule drugs remains a great challenge. Herein we describe a ligand-based interaction fingerprint (LIFt) approach for target prediction. Together with physics-based docking and sampling methods, we assessed the performance systematically by modeling the polypharmacology of 12 kinase inhibitors in three stages. First, we examined the capacity of this approach to differentiate true targets from false targets with the promiscuous binder staurosporine, based on native complex structures. Second, we performed large-scale profiling of kinase selectivity on the clinical drug sunitinib by means of computational simulation. Third, we extended the study beyond kinases by modeling the cross-inhibition of bromodomain-containing protein 4 (BRD4) for 10 well-established kinase inhibitors. On this basis, we made prospective predictions by exploring new kinase targets for the anticancer drug candidate TN-16, originally known as a colchicine site binder and microtubule disruptor. As a result, p38α was highlighted from a panel of 187 different kinases. Encouragingly, our prediction was validated by an in vitro kinase assay, which showed TN-16 as a low-micromolar p38α inhibitor. Collectively, our results suggest the promise of the LIFt approach in predicting potential targets for small-molecule drugs.