What is it about?
PFAS, often called “forever chemicals,” are widely used in industry but can harm human health and the environment because they persist for a very long time. Testing all PFAS in laboratory animals is slow, expensive, and raises ethical concerns, so there is a need for faster, alternative methods. In this study, we developed an artificial intelligence (AI) approach to predict whether a PFAS compound is highly toxic or less toxic when ingested (acute oral toxicity in rats). Using the Isalos Analytics Platform, we applied automated machine learning to compare several models and identify the most accurate one. The final model correctly classified toxicity in about 81.5% of cases.  Beyond prediction, we used a “read-across” method, which groups chemicals based on structural similarity. This allows scientists to estimate the toxicity of new or untested PFAS by comparing them to similar known compounds. The analysis showed that certain structural features—such as polyaromatic and heterocyclic elements—are linked to higher toxicity.  Importantly, this work provides a fast, reliable, and animal-free tool for screening PFAS chemicals and understanding what makes them harmful. It can support safer chemical design, regulatory decision-making, and large-scale assessment of PFAS without the need for extensive laboratory testing.
Featured Image
Photo by Ayush Kumar on Unsplash
Why is it important?
PFAS are a large and growing group of chemicals that are extremely persistent and widely detected in the environment, food, and human populations. However, only a small fraction of them have been experimentally tested for toxicity, creating a major data gap for regulators and industry. This work is important because it provides a fast, cost-effective, and animal-free approach to assess the toxicity of PFAS based on their chemical structure. By combining automated machine learning with read-across analysis, the study not only predicts toxicity with good accuracy but also helps explain which structural features drive harmful effects. The approach can support regulatory risk assessment, prioritization of hazardous substances, and the design of safer alternatives. It is particularly timely given increasing global concern and regulatory pressure around PFAS, enabling large-scale screening without the need for extensive in vivo testing.
Perspectives
This work reflects our broader effort to move toward predictive, data-driven toxicology using advanced AI tools. A key strength of this study is the integration of automated machine learning with mechanistic interpretation through read-across, allowing both accurate predictions and improved scientific understanding. From a personal perspective, the use of the Isalos Analytics Platform demonstrates how no-code/low-code tools can make sophisticated modeling approaches accessible to non-programming experts, accelerating research and decision-making. This is especially relevant for complex chemical classes like PFAS, where traditional testing cannot keep pace with the number of substances. Looking ahead, this methodology can be expanded to other toxicity endpoints and chemical families, contributing to next-generation risk assessment frameworks and supporting the transition toward more sustainable and ethical chemical evaluation practices.
Dr Antreas Afantitis
NovaMechanics Ltd
Read the Original
This page is a summary of: Read-Across Structural Analysis of PFAS Acute Oral Toxicity in Rats Powered by the Isalos Analytics Platform’s Automated Machine Learning, Toxics, February 2026, MDPI AG,
DOI: 10.3390/toxics14020152.
You can read the full text:
Contributors
The following have contributed to this page
