Adaptive one-class Gaussian processes allow accurate prioritization of oncology drug targets

Abstract Motivation The cost of drug development has dramatically increased in the last decades, with number new drugs approved per billion US dollars spent on R&D halving every year or less. selection and prioritization targets is one most influential decisions discovery. Here we present a Gaussian Process model for cast as problem learning only positive unlabeled examples. Results Since absence negative samples does not allow standard methods automatic hyperparameters, propose novel approach hyperparameter kernel One Class Processes. We compare our state-of-the-art approaches benchmark datasets then show its application to druggability prediction oncology drugs. Our score reaches an AUC 0.90 set clinical trial starting from small training 102 validated targets. recovers majority known can be used identify proteins target candidates. Availability implementation matrix features each protein available at: https://bit.ly/3iLgZTa. Source code implemented Python freely download at https://github.com/AntonioDeFalco/Adaptive-OCGP. Supplementary information data are Bioinformatics online.