Using human in vitro transcriptome analysis to build trustworthy machine learning models for prediction of animal drug toxicity

摘要: During the development of new drugs or compounds there is a requirement for preclinical trials, commonly involving animal tests, to ascertain safety compound prior human trials. Machine learning techniques could provide an in-silico alternative models assessing drug toxicity, thus reducing expensive and invasive testing during clinical that are most likely fail tests. Here we present machine model predict kidney dysfunction, as proxy induced renal in rats. To achieve this, use inexpensive transcriptomic profiles derived from cell lines after chemical treatment train our combined with structure information. Genomics data due its sparse, high-dimensional noisy nature presents significant challenges building trustworthy transparent models. address these issues by judiciously feature sets heterogenous sources coupling them measures uncertainty achieved through Gaussian Process based Bayesian We combine insight into feature-wise contributions predictions predictive uncertainties recovered improve transparency trustworthiness model.