Data Curation can Improve the Prediction Accuracy of Metabolic Intrinsic Clearance

A key consideration at the screening stages of drug discovery is in vitro metabolic stability, often measured in human liver microsomes. Computational prediction models can be built using a large quantity of experimental data available from public databases, but these databases typically contain data measured using various protocols in different laboratories, raising the issue of data quality. In this study, we retrieved the intrinsic clearance (CL_int ) measurements from an open database and performed extensive manual curation. Then, chemical descriptors were calculated using freely available software, and prediction models were built using machine learning algorithms. The models trained on the curated data showed better performance than those trained on the non-curated data and achieved performance comparable to previously published models, showing the importance of manual curation in data preparation. The curated data were made available, to make our models fully reproducible.