A scheme for the assessment and definition of tolerable uncertainty in read-across for toxicological data gap filling

The transparency and explainability of uncertainties related to read-across predictions are critical for filling toxicological data gaps. As frameworks for evaluating read-across have become standardised, so has the identification and characterisation of the various types of uncertainty, particularly those related to chemical similarity. However, it has proven more challenging to assess overall uncertainty, particularly in defining what constitutes "tolerable" uncertainty. In this study, seven areas of uncertainty related to read-across were identified and their impact on read-across for two endpoints assessed; six related to aspects of chemical structure and properties, and a further one to uncertainty within the biological data used for read-across. The impact of uncertainty associated with these seven factors was related to ordinal categories. Examples of uncertainty assessment in read-across data gap filling, where different source analogues and the same target substances were evaluated, are provided for skin sensitisation and sub-chronic systemic toxicity. The resulting scheme, a generic tabular matrix, offers a flexible and adaptable approach for assessing uncertainties related to read-across predictions, particularly those from a single-source analogue and includes an overall uncertainty level for the read-across. Analysis of existing read-across predictions provides a means to define the level of tolerable uncertainty.