Augmenting Expert Knowledge-Based Toxicity Alerts by Statistically Mined Molecular Fragments

Structural alerts are molecular substructures assumedto be associatedwith molecular initiating events in various toxic effects and an integralpart of in silico toxicology. However, alerts derivedusing the knowledge of human experts often suffer from a lack of predictivity,specificity, and satisfactory coverage. In this work, we present amethod to build hybrid QSAR models by combining expert knowledge-basedalerts and statistically mined molecular fragments. Our objectivewas to find out if the combination is better than the individual systems.Lasso regularization-based variable selection was applied on combinedsets of knowledge-based alerts and molecular fragments, but the variableelimination was only allowed to happen on the molecular fragments.We tested the concept on three toxicity end points, i.e., skin sensitization,acute Daphnia toxicity, and Ames mutagenicity, whichcovered both classification and regression problems. Results showedthe predictive performance of such hybrid models is, indeed, betterthan the models based solely on expert alerts or statistically minedfragments alone. The method also enables the discovery of activatingand mitigating/deactivating features for toxicity alerts and the identificationof new alerts, thereby reducing false positive and false negativeoutcomes commonly associated with generic alerts and alerts with poorcoverage, respectively.