Many impurities are evaluated through systemic toxicity endpoints, where acceptable exposure limits can be derived from repeat-dose studies and other toxicological data. Before such evaluations can proceed, however, the potential for mutagenicity is typically assessed. This distinction is important because impurities that interact directly with genetic material can induce mutations and are therefore managed differently from non-mutagenic impurities. As a result, hazard identification for mutagenicity becomes the primary scientific question at the outset of impurity assessment. Impurities identified as mutagenic are evaluated within the International Council for Harmonization (ICH) M7 guideline, which establishes a dedicated framework for assessing and controlling mutagenic impurities in pharmaceuticals. And those determined to be non-mutagenic may be assessed using alternative approaches.
ICH M7 Is Built on Complementary Evidence
ICH M7 recommends integrating multiple independent sources of evidence, including:
- Expert scientific review
- Existing bacterial mutagenicity data
- Carcinogenicity data where relevant
- Predicts from a statistical QSAR method
- Predictions from an expert rule-based system
Within ICH M7 assessments, existing bacterial reverse mutation (Ames) studies represent a valuable source of experimental evidence for determining mutagenic potential. However, not all Ames data carry equal weight. Greater confidence is placed in studies that evaluate the standard battery of bacterial strains recommended by OECD Test Guideline 471, and that test the substance both in the presence and absence of metabolic activation (S9).
The evidential value of a study increases when detailed information is available, including the strains tested, dose levels, metabolic activation conditions, and the final study interpretation. Original study reports with clearly documented positive, negative, or equivocal test calls generally provide stronger support for ICH M7 decision-making than literature summaries or databases containing limited information. In contrast, comprehensive data resources such as Leadscope can provide access to a richer body of genetic toxicology information, including study references, strain-specific results, metabolic activation conditions, and expert-curated test calls. Such detailed datasets enable a more robust weight-of-evidence assessment and can help contextualize findings when experimental data for the impurity of interest are limited.
Why Two QSAR Methodologies Are Required
The statistical and expert rule-based approaches provide independent lines of evidence based on different scientific principles.
Statistical models predict mutagenic potential by identifying relationships between chemical structure and experimental outcomes observed in similar compounds within a curated training dataset. In contrast, expert rule-based systems evaluate whether a compound contains structural features associated with known mechanisms of DNA reactivity and mutagenesis.
This use of complementary, independent methodologies is a defining strength of the ICH M7 framework, providing a more robust assessment than either approach could achieve alone.
Prediction Confidence Depends on Applicability Domain
The reliability of a statistical QSAR prediction depends on whether the compound falls within the model’s applicability domain.
The applicability domain defines the region of chemical space represented by the compounds used to develop and validate the model. Predictions for compounds that are well represented within this domain generally carry greater confidence because they are based on chemistry similar to that seen during model training.
An out-of-domain prediction could be viewed as an indicator of prediction confidence. For statistical models, an out-of-domain prediction suggests that the compound contains features that are underrepresented in the model’s training set, reducing the level of statistical support available for the overall prediction. However, information can still be obtained from transparent QSAR systems. For example, Leadscope enables reviewers to examine if any structural features contribute to the assessment, whether known mutagenicity-associated features are present, and whether potentially reactive features have been identified. As a result, even when a compound falls outside the formal applicability domain, the model output can, in some cases, provide additional context and help guide the broader weight-of-evidence evaluation.
This consideration becomes increasingly important as chemical discovery introduces novel chemical scaffolds that may be underrepresented in existing datasets. Applicability domains must evolve alongside emerging chemistry through the incorporation of rigorously curated experimental data, with the goal of increasing the proportion of compounds for which predictions can be made with an appropriate level of scientific confidence.
Conclusion
ICH M7 provides a scientifically robust framework for assessing the mutagenic potential of pharmaceutical impurities by integrating multiple, complementary lines of evidence. Experimental data, statistical QSAR models, expert rule-based systems, and expert review each contribute unique insights to the overall weight-of-evidence assessment.
The strength of this approach lies in the transparent interpretation of all available evidence. Comprehensive resources such as Leadscope help support this process by providing access to detailed mutagenicity data, study references, strain-specific results, and mechanistic context. Even when a compound falls outside a statistical model’s applicability domain, transparent systems can still provide valuable information about relevant structural features and potential reactivity, helping to inform expert judgement.
As pharmaceutical chemistry continues to evolve, the integration of high-quality experimental evidence, transparent computational tools, and scientific expertise will remain essential for delivering confident and defensible ICH M7 assessments.
En savoir plus
Robust ICH M7 assessments require more than individual QSAR predictions. They require integration of complementary computational methodologies, experimental evidence, applicability domain analysis, and expert scientific review.
Discover how Leadscope supports mutagenicity assessments through complementary statistical and expert rule-based models, comprehensive genetic toxicology databases, applicability domain evaluation, and workflows designed to support transparent ICH M7 assessments.
