Development of AI mutagenicity prediction system incorporating the knowledge of expert review

Abstract

We constructed the Ames mutagenicity prediction system YosAI which increases prediction accuracy as well as efficiency of mutagenicity evaluation by integrating the expert knowledge, AI technology based on internal/external Ames data and commercial prediction software, CASE Ultra.

This system can be used to (1) search for the mutagenicity and carcinogenicity risk in internal/external databases, (2) display the presence or absence of CASE Ultra structure alerts, (3) search for similar structures (fingerprint method with Tanimoto coefficient), and (4) to show the presence or absence of electrophilicity and DNA binding ability from the OECD tool box. Utilizing these in silico prediction and chemical considerations for the mechanisms of mutagenicity, we mimicked the expert reviews according to ICH M7 guideline. Furthermore, AI-based mutagenicity prediction is implemented by the Artificial Neural Network method that uses parameters such as alert, and DNA binding ability.

We hope that this prediction system can be utilized as a hybrid knowledge/statistical-SAR model for the spread of computational toxicology in pharmaceutical development.