The recent ICH M7 draft guidance states that at least two in silico models should be used for assessing potential mutagenicity of impurities, intermediates, degradants or starting materials in synthetic routes of pharmaceutical APIs.  The guidance goes further to indicate that the two models should use complementary approaches, e.g., a rules-based method and a QSAR approach.   The guidance also states that any model used for in silico evaluation should be validated according to the OECD guidelines.  In an effort to build an OECD-validated QSAR model to be used in addition to a mutagenicity predictor based on the expert system approach, we collated Ames mutagenicity data for over nine thousand compounds from several public sources.  Using in-house chemical fingerprints, an in silico mutagenicity classifier was built using a support vector machine (SVM) algorithm.  The final model was selected based on the performance of 77 available fingerprints in 50 models randomly created with 60% of data in the training set and the remaining 40% in the respective test set. The performance of the final QSAR model was evaluated on an external set of 1995 compounds assayed in Lilly Research Laboratories (LRL) for diverse drug discovery projects. With 11% indeterminate predictions, the model returned excellent negative predictive value (NPV) of over 88%, i.e., 88% of the compounds predicted to be Ames negative were confirmed so in the Ames assay. Other performance statistics of the model and its comparison with other commercial and in-house models will be presented as well as an evaluation of the importance of Lilly chemical space to model performance.

In silico model for assessment of mutagenicity to comply with ICH M7 draft guidance

246th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.