In silico prediction of ADME properties: are we making progress?

Abstract

The use of computational models in the prediction of ADME properties of compounds is growing rapidly in drug discovery as the benefits they provide in throughput and early application in drug design are realized. In addition, there is an increasing range of models available, as model builders have advanced from the first-generation' models, which were predominantly focused on solubility, absorption and metabolism, to include models of other optimization factors such as HERG, glucuronyl transferase and drug transport proteins. This widening interest is now driving demand for developments in the component elements of model building, namely higher quality datasets, better molecular descriptors and more computational power, and the quality of models is improving rapidly as a consequence. Models generally have very high throughput and can be used with virtual structures. As a consequence, they can generate large quantities of data on large numbers of compounds. Thus, one consequence of the wider choice of models, coupled with their high throughput, is a growing need to integrate their output into collective analyses of molecules against pre-set criteria. This article comments on some of the recent developments in ADME models, and highlights the importance of integrating the data to aid compound selection in drug discovery projects.