Predictive Models for Fast and Effective Profiling of Kinase Inhibitors

In this study we developed two-dimensional pharmacophore-based random forest models for the effective profiling of kinase inhibitors. One hundred seven prediction models were developed to address distinct kinases spanning over all kinase groups. Rigorous external validation demonstrates excellent virtual screening and classification potential of the predictors and, more importantly, the capacity to prioritize novel chemical scaffolds in large chemical libraries. The models built upon more diverse and more potent compounds tend to exert the highest predictive power. The analysis of ColBioS-FlavRC (Collection of Bioselective Flavonoids and Related Compounds) highlighted several potentially promiscuous derivatives with undesirable selectivity against kinases. The prediction models can be downloaded from www.chembioinf.ro .