Abstract
The Ames test is widely used to assess the mutagenic effects of chemical substances in pharmaceutical and industrial products on DNA. Positive results occur when the colony count exceeds double that of the negative control. However, the strength of mutagenicity can vary even with the same positive result. One indicator commonly used to measure mutagenicity strength in the Ames test is the specific activity value, which is the ratio of the colony count difference between the test and control divided by the dose. Higher values indicate higher mutagenicity. Chemicals with specific activity values above 1,000 are classified as having "strong mutagenicity" according to the Industrial Safety and Health Act. Recently, in silico technology that predicts mutagenicity from compound structures has been developed, enabling cost-effective and rapid evaluation. However, there is currently no software that provides information on mutagenicity strength. This makes it challenging to assess the potential for avoiding mutagenicity by modifying chemical structures after obtaining a positive result. Additionally, early access to mutagenicity severity information would facilitate compliance with legally required measures under the Industrial Safety and Health Act. To address this, we have developed a machine learning model that predicts mutagenicity strength using both in-house and publicly available data on specific activity values. We have also expanded our system to provide information on mutagenicity strength, in addition to binary determinations of negative or positive, by incorporating the model into a mutagenicity prediction system.
