Grantee Research Project Results

Development of AI-based software tool for prediction of chemical exposure and toxicity

EPA Grant Number: SU841133
Title: Development of AI-based software tool for prediction of chemical exposure and toxicity
Investigators: Wang, Qingsheng
Institution: Texas A & M University
EPA Project Officer: Callan, Richard
Phase: I
Project Period: March 1, 2025 through February 28, 2027
Project Amount: $75,000
RFA: 21st Annual P3 Awards: A National Student Design Competition Focusing on People, Prosperity, and the Planet Phase I (2024) RFA Text
Research Category: Pesticides , Human Health , P3 Awards , P3 Challenge Area - Chemical Safety , Chemical Safety for Sustainability , Environment , Predictive Toxicology

Description:

Chemical safety is essential for the well-being of individuals, communities, economies, and the environment. Our interdisciplinary student team will develop an innovative technology to accurately identify and evaluate the toxicity/exposure of new chemical substances such as pesticides, enabling appropriate control and uninterrupted operation of facilities, business, and industries. This sustainable approach will protect the environment, strengthen communities, and generate economic benefits for the U.S. Additionally, a user-friendly software tool will be developed to disseminate this technology to potential users across the U.S. This project will be integrated into the current curriculum of chemistry, chemical engineering and public health at Texas A&M University and Prairie View A&M University, providing educational opportunities for diverse communities in partnership with a historically black university.

Objective:

This project aims to address EPA-2024-P3-Q4 funding opportunity of Objective 7.1: Ensure Chemical and Pesticide Safety (Goal 7: Ensure Safety of Chemicals for People and the Environment). Traditionally, investigating the exposure and toxicity of new chemicals or their mixtures in the lab has required a significant amount of complex and costly animal testing. Similarly, the widely used methods for evaluating toxicity in chemical mixtures, such as dose and response addition, do not consider interactions between substances. Instead, we propose to use current AI techniques and existing research data to develop a software tool for efficient exposure and toxicity prediction. The challenges include identifying the best techniques and conducting comprehensive literature reviews to collect sufficient data for training the prediction models. The resulting software tool will be user-friendly and easily accessible on the website, making it convenient for federal agencies like the U.S. EPA, as well as industries, to implement the software in their decision-making processes. With this innovative tool, first responders, communities, facilities, and authorities will be better equipped to plan appropriate response procedures in the event of a chemical release incident. Additionally, this work aligns with United Nations' Sustainable Development Goal 11, to make cities and human settlements inclusive, safe, resilient, and sustainable by improving understanding of chemical exposure and toxicity.

 

Expected Results:

A comprehensive database of chemical exposure and toxicity will be created through literature reviews and data searches. A Graph Neural Network (GNN) with autoencoder will be applied to encode molecular geometric information for better understanding of exposure and toxicity in chemical substances. Quantitative Structure-Activity Relationship (QSAR) models will then be developed to predict toxicity in chemicals and mixtures based on the molecular information from GNN, along with a software tool for these models. Statistical assessment will be applied to evaluate the model performance. Demonstrations of the use of this software tool will be conducted with first responders and industries to ensure all potential users can make informed decisions about the thousands of chemicals used in the U.S. to enhance chemical safety.

Supplemental Keywords:

Interactive development tools; computer models; toxic use reduction