Assessing Uncertainty in Read-across: Questions to Evaluate Toxicity Predictions Based on Knowledge Gained from Case Studies
Schultz, T. W.; Richarz, A.-N.; Cronin, M. T. D. Computational Toxicology 2019, 9, 1–11.
Read-across as an alternative assessment method for chemical toxicity has growing interest in both the regulatory and industrial communities. The pivotal means of acquiring acceptance of a read-across prediction is identifying and assessing uncertainties associated with it. This study has identified and summarised, in a structured way, the variety of uncertainties that potentially impact acceptance of a read-across argument. The main sources of uncertainty were established and divided into four main categories: i) the regulatory use of the prediction, ii) the data for the apical endpoint being assessed, iii) the read-across argumentation, and iv) the similarity justification. Specifically, the context of, and relevance to, the regulatory use of a read-across will dictate the acceptable level of uncertainties. The apical endpoint (or other) data must be of sufficient quality and relevance for data gap filling. Read-across argumentation uncertainties include: i) mechanistic plausibility (i.e., the knowledge of the chemical and biological mechanisms leading to toxicity), ii) completeness of the supporting evidence, iii) robustness of the supporting data, and iv) Weight-of-Evidence. In addition, similarity arguments for chemistry, physico-chemical properties, toxicokinetics and toxicodynamics are linked to these read-across argumentation issues. To further progress in this area, a series of questions are proposed with the goal of addressing each type of uncertainty.