DEVELOPMENT AND APPLICATIONS OF GLOBAL ADMET MODELS
Karl‐Heinz Baringhaus, Gerhard Heßler, Hans Peter Matter, Friedemann Schmidt · 2013
Absorption, distribution, metabolism, excretion, and toxicology (ADMET) strongly influence the pharmacokinetic and safety behavior of drugs. Therefore, ADMET parameters determined by in vitro assays are used in drug discovery to prioritize hit and lead structures as well as to support compound optimization. The large datasets obtained from these efforts provide a good opportunity to develop global in silico models, which can successfully be used in the selection of hit or lead structure with as few liabilities as possible. Such models, if carefully validated, offer additional opportunities in lead optimization. Here, we describe the development of a global model for metabolic lability from a large in-house human liver microsome dataset. The model building strategy comprised the evaluation of different molecular descriptors and decision tree enhancements provided by the software Cubist to identify the machine learning algorithm and the descriptors performing best for this dataset. To ensure proper application of the model, a similarity-based applicability domain estimation was implemented, together with a continuous model update procedure. This combination ensures reliable prediction of metabolic lability, which can be used along the early drug discovery process. A case study is given, illustrating the application of the model in a multidimensional compound optimization.