The importance of optimization of ADMET properties within the drug design process is undeniable. The earlier the ADMET profiling is introduced, the better, as even the most active compounds can be rejected from further research due to the unfavorable physicochemical or pharmacokinetic properties or due to the induction of toxic effects.

In the study, we focused on metabolic stability, which is important not only due to the fact that the candidate drug should be stable enough to have sufficient time to produce therapeutic effect, but also due to the possibility of the formation of toxic products after transformations. We developed a tool, which enables not only in silico evaluation of metabolic stability (expressed as compound half-lifetime), but also provides pieces of information, which can guide the process of structure optimization with the reference to the evaluated parameter.

The compound half-lifetime is assessed with the use of machine learning algorithms developed on the respective data from the ChEMBL database, for compounds represented with different substructural fingerprints. Shapley Additive exPlanations (SHAP) theory was applied to examine the influence of the particular chemical moieties on the model’s outcome. We also prepared the web service (available at https://metstab-shap.matinf.uj.edu.pl/), where compounds submitted by the user can be evaluated in terms of the contribution of particular structural features to the outcome of half-lifetime predictions.

Acknowledgments

The study was supported by the grant OPUS 2018/31/B/NZ2/00165 financed by the National Science Centre, Poland (www.ncn.gov.pl)