”Parallelized variable selection and modeling based on prediction” algorithm on GPU for feature selection and ADMET model generation

Lack of safety and efficacy are the two major reasons for the failures of drug candidates in drug discovery and development. Reliable prediction of blood-brain barrier permeation even before chemical synthesis still remains as one of the major challenges in drug discovery. New approaches and models that are reliable and can reduce experimental evaluations of pre-clinical candidates are in urgent need. We describe herein Parallelized Variable Selection and Modeling based on the Prediction, an exhaustive search algorithm for feature selection and robust model ADMET model generation using publically available blood-brain barrier permeation data.