Pobody’s Nerfect: (Q)SAR works well for predicting bacterial mutagenicity of pesticides and their metabolites, but predictions for clastogenicity in vitro have room for improvement

Abstract

Genotoxicity assessment is a key component of regulatory decision-making in pesticide authorization and biocide approval. Conventionally, these genotoxicity requirements are addressed with OECD test guideline-compliant in vitro tests. In recent years, in silico approaches, such as (Q)SAR, have matured sufficiently so that they may be suitable to support, complement or even replace in vitro testing as a first tier of genotoxicity assessment. Among the different endpoints for genotoxicity, a high reliability is expected for in silico predictions of the endpoint bacterial mutagenicity. For other endpoints predictive performance is either unclarified or seems to be comparably lower. Herein, we describe the evaluation of several commercial and freely available (Q)SAR models and complementary combinations thereof with respect to the endpoints bacterial mutagenicity and chromosome damage in vitro. We used curated in-house test sets derived from OECD test guideline-compliant studies, gathered from submissions for the regulatory approval of biocides and plant protection products. The data set comprises active substances, metabolites and impurities. In line with previous publications we show that (Q)SAR models for bacterial mutagenicity generally performed well for compounds of the pesticide domain. Model combinations significantly increased the respective sensitivity. Models for chromosome damage still need to improve prior to their stand-alone use in regulatory decision-making, either strongly leaning towards sensitivity, at the expense of specificity or vice versa. Similar to the endpoint bacterial mutagenicity, combinations of models for chromosome damage increase sensitivity when compared to the individual models alone.