Procedure for toxicological predictions based on mechanistic weight of evidences: Application to Ames mutagenicity
Computational Toxicology Volume 12 (2019) 100109
Typically, (Q)SAR models are deemed decision-making rather than decision-supporting computational methods. In some (Q)SARs, the relation between chemicals structures and biological activity is described statistically. In other models, this relation is based on mechanistic-related events which contribute to the apical effect. Whether the definitive decision is based on a single model or series of models, usually predictions are not supported by mechanistic justification. The lack of mechanistic justification often limits the use of (Q)SAR predictions, especially for regulatory decisions. With this in mind, a workflow based on combining mechanistic (Q)SAR, read-across analysis and expert knowledge is used examine four different scenarios where the workflow provides enough weight-of-evidence, to allow users to make a transparent decision as to the ultimate prediction. When the OASIS TIMES Ames model and read-across analysis based on well-selected analogues within the OECD Toolbox show consistent predictions, expert input may not be needed to make a final decision. Nonetheless, expert input may be useful by adding weight-of-evidence by expanding the set of read-across analogues from literature sources and/or for providing rational of the endpoint-specific similarity between source analogue(s) and target chemical. In cases where there is inconsistency between TIMES Ames model and read-across predictions, expert input is critical for assigning the ultimate effect. Specifically, expert evaluation assists in assessing the correctness/validity of used experimental data, as well as assessing their recentness, presence of S9 metabolic activation accordance to guideline protocols, cytotoxicity, etc. In the latter cases, after evaluating all relevant evidences, expert knowledge provides transparency for the conclusion regarding the ultimate effect.