SafetAI Initiative
To develop novel deep learning methods for toxicological endpoints that are critical to the safety review of drug candidates before entering clinical trials.
![SafetAI Logo](https://public4.pagefreezer.com:443/content/FDA/18-08-2023T10:34/https://www.fda.gov/files/styles/main_image_medium/public/safetAI.png?itok=BnT2-6Y8)
Objective: To develop AI models for toxicological endpoints that are critical to assess drug safety and may add value to the review of drug candidates prior to human testing.
Introduction: Drug safety is of great concern to public health. In addition, during the Investigational New Drug (IND) application submission process, the FDA specifically reviews the safety of the submitted drug candidate before the sponsor can initiate any clinical trials. SafetAI is a collaborative initiative led by CDER (e.g., funded by a FY2022 CDER Safety Research Interest Group grant), where NCTR is developing a suite of deep learning based QSAR models for various safety endpoints critical to regulatory science and the IND review. Currently, the initiative has focused on five key safety endpoints: hepatotoxicity, carcinogenicity, mutagenicity, nephrotoxicity, and cardiotoxicity.
Approaches: We are developing a deep learning framework which is designed to optimize toxicity prediction for individual chemicals based on their chemical characteristics. The method was compared to several conventional machine learning and state-of-art deep learning methods for predicting drug-induced liver injury (i.e., DeepDILI), carcinogenicity (i.e., DeepCarc), and Ames mutagenicity (i.e., DeepAmes). The preliminary results from this method yielded significant improvement in these toxicity endpoints in comparison to other deep learning and QSAR methods.
Potential Impact: SafetAI facilitates drug safety research with a novel deep learning architecture that improves the “precision” in toxicity assessment by tailoring prediction to chemical characteristics. It could play a role in providing critical safety information during the IND review process.
References:
- DeepDILI: Deep Learning-Powered Drug-Induced Liver Injury Prediction Using Model-Level Representation.
Li T., Tong W., Roberts R., et al.
Chemical Research in Toxicology. 2021, 34:550-565.
- Deep Learning on High-Throughput Transcriptomics to Predict Drug-Induced Liver Injury.
Li T., Tong W., Roberts R., et al.
Frontiers in Bioengineering and Biotechnology. 2020, 8.
- DeepCarc: Deep Learning-Powered Carcinogenicity Prediction Using Model-Level Representation.
Li T., Tong W., Roberts R., et al.
Frontiers in Artificial Intelligence. 2021, 4.