Prediction of In Vivo Pharmacokinetic Parameters and Time–Exposure Curves in Rats Using Machine Learning from the Chemical Structure

• DOI: 10.1021/acs.molpharmaceut.2c00027
• Publication Date: 2022-04-12T13:16:26Z
• Authors: Olga Obrezanova; Anton Martinsson; Tom Whitehead; Samar Mahmoud; Andreas Bender; Filip Miljković; Piotr Grabowski; Ben Irwin; Ioana Oprisiu; Gareth Conduit; Matthew Segall; Graham F. Smith; Beth Williamson; Susanne Winiwarter; Nigel Greene
• Keywords: ADME

Animal pharmacokinetic (PK) data as well human and animal in vitro systems are utilized drug discovery to define the rate route of elimination. Accurate prediction mechanistic understanding clearance disposition animals provide a degree confidence for extrapolation humans. In addition, vivo properties can be used improve design during discovery, help select compounds with better properties, reduce number experiments. this study, we generated machine learning models able predict rat PK parameters concentration-time profiles based on molecular chemical structure either measured or predicted parameters. The were trained internal over 3000 diverse from multiple projects therapeutic areas, endpoints include oral bioavailability. We compared performance various traditional algorithms deep approaches, including graph convolutional neural networks. best achieved R2 = 0.63 [root mean squared error (RMSE) 0.26] 0.55 (RMSE 0.46) fast cost-efficient way guide molecules optimal profiles, enable virtual at point design, drive prioritization assays.