Abstract
It is not unusual for models developed to predict Absorption, Distribution, Metabolism, Elimination, and Toxicity (ADMET) as well as other endpoints to be nothing more than a black box, supplying only a numerical or categorical value by which users are expected to assess the goodness of a query compound. This type of result is, however, of little use in the overarching goal of developing better and safer drug compounds, that is, providing guidance toward improving the characteristics of the query or formulating chemical hypotheses that can be evaluated through synthetic efforts. As a result, the level of acceptance of predictions by users outside the computational chemistry and molecular modeling groups tends to be very low. In this review, I address three primary domains in which model presentation can be improved, specifically, establishing confidence in the prediction, data interrogation, and model interpretability. It has been my experience that even small efforts in these areas result in a much greater return with respect to acceptance, good faith, and usage from users regardless of the users role within the drug discovery process.
3