High-accuracy prediction of mechanisms of action using structural alerts

Abstract

Knowing the mechanism of action (MechoA) of substances is a crucial first step in determining an Adverse Outcome Pathway and in risk assessment, especially when using in silico models to predict (eco)toxicity. We developed a set of structural alerts associated with specific MechoAs based on (eco)toxicity data on hundreds of chemicals and compiled them into a new method to predict MechoAs with high accuracy and with simple rules. The MechoA scheme classifies substances into 6 general MechoAs and 23 detailed MechoA sub-groups. The rules are mainly based on data on mammals and fish. We used a training set of 301 chemicals, and a validation set of 493 molecules. We achieved with this method 91.4% correct classifications for the training set and 92.2% for the validation set. This model is both simpler and performs better than the previous (quantic chemistry based) model we developed and we recommend its use for AOP compilation and for risk assessment. This model will be continuously enhanced with the addition of new rules and minor corrections as they are discovered.