top of page

Insights from the the Second Ames/QSAR international challenge project


Quantitative Structure-Activity Relationship (QSAR) models serve as invaluable in silico tools for predicting mutagenicity, particularly for compounds challenging to assess through traditional methods like the Ames test. The quest for robust regulatory models led to the Division of Genetics and Mutagenesis at the National Institute of Health Sciences, Japan (DGM/NIHS), orchestrating the Second Ames/QSAR International Challenge Project (2020–2022).


"Evaluation of QSAR models for predicting mutagenicity: outcome of the Second Ames/QSAR international challenge project" has been recently published as an open access paper.


Regulatory bodies are increasingly turning to in silico methods as a strategic approach to address critical issues related to animal welfare, cost reduction, and the evaluation of chemicals that pose challenges for traditional in vivo and in vitro tests. In this context, quantitative structure−activity relationship (QSAR) models emerge as powerful tools, capable of predicting the biological activities of chemicals based on their molecular structures.


A specific category within QSAR models, the Ames/QSAR, derives its foundation from Ames test data and excels in predicting the mutagenicity of various chemicals. The application of Ames/QSAR models extends to the evaluation of impurities in pharmaceuticals and diverse chemicals, including pesticides and their metabolites. Notably, existing Ames/QSAR models, often constructed from publicly available datasets, exhibit high accuracy in predicting the mutagenicity of known chemicals in the public domain.


However, a significant challenge arises when these models are confronted with the detection of new Ames-positive compounds. The inherent limitations stem from imbalanced datasets, with unequal representation of Ames-positive and -negative chemicals, contributing to decreased model sensitivity. Although balanced accuracy may offer a better measure of performance against unbalanced proprietary datasets, the primary obstacle lies in the insufficient coverage of chemical space and mutagenic mechanisms within the training sets or expert rules.


To address these limitations and enhance the regulatory utility of Ames/QSAR models, the Division of Genetics and Mutagenesis at the National Institute of Health Sciences, Japan (DGM/NIHS), initiated the First Ames/QSAR International Challenge Project (2014–2017). This pioneering effort involved 12 teams, predominantly comprising QSAR model vendors, from seven countries. Participants utilized their Ames/QSAR models to predict the mutagenicity of approximately 12,000 new chemicals, leading to significant improvements in the predictive ability of these models.


Building upon the success and insights gained from the First Project, the DGM/NIHS conducted the Second Ames/QSAR International Challenge Project (2020–2022). Noteworthy changes included broader participation, involving more academic and non-commercial entities, as well as the incorporation of both deep-learning and conventional QSAR models. The training dataset remained consistent with that of the First Project, and a new trial dataset comprising 1,589 chemicals was introduced. The Second Project aimed to further refine Ames/QSAR models by providing curated training data, mutagenicity results for multiple test strains, information on metabolic activation, solvents used, and test chemical purity. While the models demonstrated high specificity but low sensitivity compared to the First Project, the experience proved invaluable. Notably, the nine teams participating in both projects exhibited improved sensitivity. Learn more about the outcomes and insights gained in the full conclusion.




58 views0 comments

Commentaires


bottom of page