A new classification algorithm based on mechanisms of action

Abstract

A good understanding of Mechanisms of Action (MechoAs) and appropriate methods to determine them is crucial for the accurate prediction of toxicity using in silico techniques. Different MechoAs can be related to different QSAR models to predict toxicity values. Therefore, we defined a set of MechoAs, based on Molecular Initiating Events, the first step of Adverse Outcome Pathways (AOPs). In the most common classification algorithms used to predict Modes of toxic Action, the prediction domain is often limited by the need to identify known structural alerts. To circumvent this limitation, we developed a new algorithm to predict MechoAs principally based on mammals and fish, using molecular modelling to obtain calculated molecular parameters. Comparing the Verhaar scheme (as modified by Enoch et al. (2008)) with the MechoA method for the same validation set, MechoA achieved 69% correct classifications as opposed to 45% for the Verhaar scheme, 17% misclassifications for both, 13% classifications slightly different from the literature for our algorithm. No substances fell into zones where two possible MechoAs couldn’t be differentiated from each other, while 1% of the molecules were out of the prediction domain of the algorithm as opposed to 38% using the Verhaar scheme. Thus, this model enhances precision of correct AOP identification for in silico toxicity predictions.