Texas A&M develops artificial intelligence tools to assess chemical safety
Scientists at Texas A&M University have developed artificial intelligence tools designed to estimate both the potential hazards of chemicals and the reliability of those predictions. Led by Dr. Weihsueh Chiu from the Department of Veterinary Physiology and Pharmacology, the research expands on previous findings published in Nature Communications regarding quantitative structure-activity relationship models. These machine learning models utilize chemical structures to estimate safe exposure levels, moving beyond traditional evaluations that rely heavily on animal studies or long-term human epidemiological research. The primary goal is to address the significant gap in comprehensive safety data for the large number of chemicals currently in commerce. By incorporating uncertainty-aware machine learning, the models can estimate how confident they are in each prediction based on the quality and quantity of existing data for similar substances. This transparency allows researchers to identify chemicals that may require additional study or expert review, particularly those with limited supporting data. When applied to over 126,000 chemicals, the models identified patterns in both toxicity and uncertainty. Substances such as metals, polychlorinated compounds, and per- and polyfluoroalkyl substances (PFAS) frequently exhibited higher uncertainty levels due to complex behavior or data scarcity. The researchers propose a tiered evaluation process where AI handles large-scale screening while experts focus on high-risk or uncertain substances. Continued advances in this technology aim to shift chemical safety evaluation from a reactive process toward a more predictive approach. This methodology supports regulators and scientists in identifying potential hazards more efficiently, ultimately improving public health outcomes through better risk assessment. The core advancement lies in the ability of these AI models to quantify prediction uncertainty alongside toxicity estimates, offering a more transparent view of chemical risks. This capability is significant because it helps prioritize testing efforts toward areas where scientific knowledge is currently limited, such as with PFAS and certain metal compounds. While the technology promises a shift toward predictive safety assessments, its effectiveness depends on the continued accumulation of high-quality data for diverse chemical classes. Future implementation will likely require a hybrid approach combining algorithmic screening with human expert oversight to manage residual uncertainties.
Anzeigenöffentlicht: June 4, 2026 at 12:21 AM
News Article
Inhalt
Scientists at Texas A&M University have developed artificial intelligence tools designed to estimate both the potential hazards of chemicals and the reliability of those predictions. Led by Dr. Weihsueh Chiu from the Department of Veterinary Physiology and Pharmacology, the research expands on previous findings published in Nature Communications regarding quantitative structure-activity relationship models. These machine learning models utilize chemical structures to estimate safe exposure levels, moving beyond traditional evaluations that rely heavily on animal studies or long-term human epidemiological research.
The primary goal is to address the significant gap in comprehensive safety data for the large number of chemicals currently in commerce. By incorporating uncertainty-aware machine learning, the models can estimate how confident they are in each prediction based on the quality and quantity of existing data for similar substances. This transparency allows researchers to identify chemicals that may require additional study or expert review, particularly those with limited supporting data.
When applied to over 126,000 chemicals, the models identified patterns in both toxicity and uncertainty. Substances such as metals, polychlorinated compounds, and per- and polyfluoroalkyl substances (PFAS) frequently exhibited higher uncertainty levels due to complex behavior or data scarcity. The researchers propose a tiered evaluation process where AI handles large-scale screening while experts focus on high-risk or uncertain substances.
Continued advances in this technology aim to shift chemical safety evaluation from a reactive process toward a more predictive approach. This methodology supports regulators and scientists in identifying potential hazards more efficiently, ultimately improving public health outcomes through better risk assessment.
Wichtige Erkenntnisse
The core advancement lies in the ability of these AI models to quantify prediction uncertainty alongside toxicity estimates, offering a more transparent view of chemical risks.
This capability is significant because it helps prioritize testing efforts toward areas where scientific knowledge is currently limited, such as with PFAS and certain metal compounds.
While the technology promises a shift toward predictive safety assessments, its effectiveness depends on the continued accumulation of high-quality data for diverse chemical classes.
Future implementation will likely require a hybrid approach combining algorithmic screening with human expert oversight to manage residual uncertainties.
Redaktionsauswahl
Keine Produkte verfügbar