Characterisation of data resources for in silico modelling: benchmark datasets for ADME properties

The cost of in vivo and in vitro screening of ADME properties of compounds has motivated efforts to develop a range of in silico models. At the heart of the development of any computational model are the data; high quality data are essential for developing robust and accurate models. The characteristics of a dataset, such as its availability, size, format and type of chemical identifiers used, influence the modelability of the data. Areas covered: This review explores the usefulness of publicly available ADME datasets for researchers to use in the development of predictive models. More than 140 ADME datasets were collated from publicly available resources and the modelability of 31 selected datasets were assessed using specific criteria derived in this study. Expert opinion: Publicly available datasets differ significantly in information content and presentation. From a modelling perspective, datasets should be of adequate size, available in a user-friendly format with all chemical structures associated with one or more chemical identifiers suitable for automated processing (e.g. CAS number, SMILES string or InChIKey). Recommendations for assessing dataset suitability for modelling and publishing data in an appropriate format are discussed.