The Office of Food Additive Safety (OFAS) at US FDA’s Center for Food Safety and Applied Nutrition (CFSAN) is responsible for ensuring the safety of all food additives used in the United States. A database called Chemical Evaluation and Risk Estimation System (CERES) was created for reviewers at OFAS to serve as a knowledgebase of chemical, toxicological, and regulatory information on food additives, food contact substances, and GRAS ingredients. Over time, the CERES database was enhanced with data on other chemicals such as colors, cosmetic ingredients, and drugs along with cheminformatics tools such as similarity searching and toxicity prediction. One of our current research efforts at OFAS focuses on building in-house mutagenicity and carcinogenicity predictive models with high prediction accuracy for food related chemicals. In this research, we present an evaluation of different chemical fingerprinting methods and machine learning algorithms available in the public domain and compare their performance for <i>in silico</i> prediction of Ames mutagenicity using the benchmark dataset (Hansen et al. 2009). We evaluate six fingerprinting methods, namely MACCS keys, PubChem fingerprints,  Atom pairs, Topological torsions, RDKit fingerprints, and Circular fingerprints and six machine learning algorithms, namely Naïve Bayes (NB), k-nearest neighbors (kNN), Decision Trees (DT), Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN). Using 5-fold cross-validation scheme, we calculate sensitivity, specificity, and concordance for all combinations of fingerprinting methods and machine learning algorithms and identify models that give high performance metrics. The predictive performance of these mutagenicity models is also evaluated against empirical data within CERES. In addition, the aforementioned tools are used to evaluate the chemical space of Hansen dataset and compare it to that of the food related chemicals within CERES.

Evaluating performance of chemical fingerprinting methods and machine learning algorithms for in silico prediction of Ames mutagenicity

253rd National Meeting (2017)

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