Structure–Activity Relationship of Protein-Mediated Molecular Initiating Events of Organ-Specific Toxicities

Toxicological research is facing a major transition from animal models into <i>in vitro</i> and <i>in silico</i> models to improve the cost-effectiveness of the testing process, shorten the timeline for primary screening of chemicals, and better align with the 3Rs principles to reduce animal suffering. In this work, structure-activity relationships were developed based on structural alerts (SAs) that flag the ability of chemicals to trigger specific molecular initiating events (MIEs) upstream of five adversities (cholestasis, steatosis, kidney tubular necrosis, cognitive functional defects, and neural tube closure defects). Twenty-nine protein targets linked to MIEs were identified from published adverse outcome pathway networks, while bioactivity data for chemicals against such targets were collected from ChEMBL 35 database. SARpy 2.0 included in the novel ΔQSAR (DeltaQSAR) software was used to extract rulesets, i.e., collections of structural alerts codifying for protein bioactivity. The rulesets were evaluated using an external test set to assess their real-life predictivity. Good external validation performance was achieved for 22 out of 29 rulesets that returned balanced accuracy ≥ 70% and coverage ≥ 70%, confirming that these rulesets can be used for high-throughput as well as preliminary testing of chemicals. Moreover, combining structure-activity relationship with the adverse outcome pathway concept provides a mechanistic basis to the prediction suggested by the rulesets.