Introduction

The “Ames test”, originally developed by Bruce Ames and his group, is a way to measure the mutagenic potential of chemicals.1 It uses strains of Salmonella typhimurium and Escherichia coli as an alternative to chronic dosing in rodents. It is a short-term bacterial reverse mutation assay that detects the type of chemically-induced genetic damage that could be carcinogenic in humans. For impurities below the ICH2 qualification thresholds, and lacking data for classification with respect to mutagenic and carcinogenic potential, the regulatory bodies of the European Union, Japan, and the USA accept genotoxicity evaluation by (Q)SAR approaches. Here we present a set of statistically-based QSAR models that predict outcomes of the bacterial reverse mutation assay.

Ten models were created from data on five individual strains of S. typhimurium or E. Coli, with and without rat liver S9 metabolic activation. Artificial neural network ensemble (ANNE) classification methodology was used to train the models.3 Based on predictions from these models, rules were developed to assess mutagenicity risk (MUT_Risk) using a focused subset of 2,270 compounds from the World Drug Index4 (WDI). For a subset of the Hansen’s benchmark test set5 that was not included in the training sets of our models, it was observed that 94.2% (114/121) of the compounds with a MUT_Risk of 4 were, in fact, mutagenic.

SOT 55th Annual Meeting and ToxExpo, March 13-17, 2016, New Orleans, LA

By Michael Lawless, Vijay K. Gombar, and Robert D. Clark