Computational approaches to DMPK: A realistic assessment of current methods and their practical impact. Part I: Physicochemical and in vitro properties

• Discussion of predictive models for physicochemical and in vitro absorption, distribution, metabolism, and excretion (ADME) to support decision-making. • Emphasis of the importance of not only tech-push, but also science-pull approaches. • Highlights experimental error, data volume, and threshold criteria as key factors for better predictive models. • Stresses that strong collaboration between experimental biologists and machine learning researchers is essential for success. Artificial intelligence and computational approaches have received considerable interest in recent years, and here we assess their real-world utility in drug discovery projects. We review recent in silico models in the area of drug metabolism and pharmacokinetics (DMPK), especially for physicochemical properties (pKa and logD) and in vitro assays [solubility (DMSO, Dried-DMSO, Powder), permeability (PAMPA, Caco-2, MDCK), metabolic stability (liver microsome, hepatocyte), and protein binding (plasma, microsome, brain)]. We discuss which are currently fit for purpose (and which are not), bridging both computational and experimental aspects in the early drug discovery stages. The review includes diverse aspects of obtaining data and model generation, as well as modeler/experimentalist interplay.