The publication of the International Council for Harmonisation (ICH) M7 Guideline in 2014 was groundbreaking for the field of computational toxicology because it was the first internationally harmonized testing guideline to suggest the use of (Q)SAR in a testing strategy. The guideline, which has been updated since its initial adoption, suggests that to reduce potential carcinogenic risk in pharmaceuticals, computational approaches can be used to predict the outcome of a bacterial mutagenicity assay to identify and classify mutagenic impurities1. According to the guidance,
“A computational toxicology assessment should be performed using (Q)SAR methodologies that predict the outcome of a bacterial mutagenicity assay (Ref. 6). Two (Q)SAR prediction methodologies that complement each other should be applied. One methodology should be expert rule-based and the second methodology should be statistical-based. (Q)SAR models utilizing these prediction methodologies should follow the general validation principles set forth by the Organisation for Economic Co-operation and Development (OECD).”
To identify and classify impurities, the guidance encourages the use of two (Q)SAR methodologies: expert rule-based and statistical. An expert rule-based model makes predictions about a chemical’s toxicity based on structure-activity relationships (SAR), a set of rules derived from expert knowledge that relate the structural features of a molecule with a specific biological activity. For ICH M7, expert rule-based models are used to identify specific structural features in a query chemical that are associated with bacterial mutagenicity (those published by Ashby and Tennnant2, for example). The GT_EXPERT model in CASE Ultra is uses expert rule-based methodology to predict bacterial mutagenicity for ICH M7 evaluation.
In contrast, a statistical model quantitatively relates a chemical’s structure to a toxicological endpoint using quantitative structure-activity relationships (QSAR). By analyzing a dataset with known toxicological activity, the model will identify statistically significant structural features of active and inactive compounds in the dataset. This method aims to quantify the relationship between the compound’s structure and the biological activity without the use of human knowledge using computational techniques such as linear regression and machine learning3. Not only can statistical models capture features that highly correlate to mutagenicity, these models also have the ability to detect mitigating features that reduce the likelihood of a compound’s ability to induce mutation4. GT1_BMUT is the statistical model used within CASE Ultra to predict bacterial mutagenicity for ICH M7 assessments.
The use of both methodologies, collectively referred to as (Q)SAR, is recommended in the ICH M7 guideline5. This practice incorporates the strengths of each methodology into the assessment, improving the predictive performance and providing increased confidence in prediction results6. Like experiments that take place in a wet lab, different in silico models that predict for the same endpoint and are validated under OECD principles have the potential to yield different prediction results due to underlying differences in factors like methodology, choice of algorithm, and training data.
(1) ICH. M7(R2) Assessment and Control of DNA Reactive (Mutagenic) Impurities in Pharmaceuticals to Limit Potential Carcinogenic Risk. https://database.ich.org/sites/default/files/ICH_M7%28R2%29_Guideline_Step4_2023_0216_0.pdf.
(2) Ashby, J.; Tennant, R. W. Definitive Relationships among Chemical Structure, Carcinogenicity and Mutagenicity for 301 Chemicals Tested by the U.S. NTP. Mutat. Res. Genet. Toxicol. 1991, 257 (3), 229–306. https://doi.org/10.1016/0165-1110(91)90003-E.
(3) Gini, G. QSAR Methods. In In Silico Methods for Predicting Drug Toxicity; Benfenati, E., Ed.; Springer US: New York, NY, 2022; pp 1–26. https://doi.org/10.1007/978-1-0716-1960-5_1.
(4) Jayasekara, P. S.; Skanchy, S. K.; Kim, M. T.; Kumaran, G.; Mugabe, B. E.; Woodard, L. E.; Yang, J.; Zych, A. J.; Kruhlak, N. L. Assessing the Impact of Expert Knowledge on ICH M7 (Q)SAR Predictions. Is Expert Review Still Needed? Regul. Toxicol. Pharmacol. 2021, 125, 105006. https://doi.org/10.1016/j.yrtph.2021.105006.
(5) Sutter, A.; Amberg, A.; Boyer, S.; Brigo, A.; Contrera, J. F.; Custer, L. L.; Dobo, K. L.; Gervais, V.; Glowienke, S.; Gompel, J. van; Greene, N.; Muster, W.; Nicolette, J.; Reddy, M. V.; Thybaud, V.; Vock, E.; White, A. T.; Müller, L. Use of in Silico Systems and Expert Knowledge for Structure-Based Assessment of Potentially Mutagenic Impurities. Regul. Toxicol. Pharmacol. 2013, 67 (1), 39–52. https://doi.org/10.1016/j.yrtph.2013.05.001.
(6) Kruhlak, N. L.; Benz, R. D.; Zhou, H.; Colatsky, T. J. (Q)SAR Modeling and Safety Assessment in Regulatory Review. Clin. Pharmacol. Ther. 2012, 91 (3), 529–534. https://doi.org/10.1038/clpt.2011.300.