By utilizing artificial intelligence, researchers at Chalmers University of Technology and the University of Gothenburg in Sweden have developed a method to enhance the identification of toxic chemicals based solely on their molecular structure. This AI method can assist in better regulating the growing number of chemicals used in society, while also reducing reliance on animal testing. Chemicals are prevalent in various products and processes, posing risks to humans and the environment when they enter waterways. PFAS, a group of problematic substances, has been found in alarming concentrations in groundwater and drinking water, highlighting the need for effective toxicity assessments.
Despite existing chemical regulations, negative effects persist, necessitating time-consuming animal testing to establish safety standards. With over two million animals used annually in the EU alone for regulatory compliance, the rapid development of new chemicals presents challenges in determining toxicity. The AI method developed by Swedish researchers offers a cost-effective solution for assessing chemical toxicity early on, minimizing the need for extensive animal testing. By training on large datasets from past laboratory tests, the method accurately predicts toxic properties of previously untested chemicals based on their chemical structures.
With over 100,000 chemicals on the market, a majority lack well-defined toxicity profiles, making comprehensive toxicity assessments impractical through conventional methods like animal testing. The researchers believe that their AI method offers a viable alternative for assessing the toxicity of diverse chemicals, benefitting environmental research, regulatory bodies, and chemical developers. By making the method publicly available, they aim to facilitate its use in a wider context and promote the development of safer chemicals to mitigate the negative impacts of chemical pollution on human health and ecosystem services.
Comparing their AI method with existing computational tools, the researchers found that their approach exhibits higher accuracy and broader applicability. Drawing on advanced deep learning techniques, the AI model based on transformers leverages vast datasets to enhance predictions of chemical toxicity. As AI-based methods continue to evolve with increasing data availability, the researchers anticipate their potential to outperform conventional computational approaches, potentially replacing laboratory tests in toxicity assessments. This transition could not only reduce animal experimentation but also streamline the development of new chemicals and identify safer alternatives to toxic substances currently in use.
The AI method’s foundation in transformers, originally developed for language processing, demonstrates its versatility in capturing information from chemical structures. By identifying specific properties within molecules that contribute to toxicity, transformers enable more sophisticated toxicity predictions through deep neural networks. This AI framework continuously enhances its predictive capabilities through training on extensive data from prior laboratory experiments on the effects of thousands of chemicals on various organisms. By harnessing transformers and neural networks, the AI method offers a promising avenue for advancing computational assessments of chemical toxicity, potentially revolutionizing the regulatory landscape and fostering the development of safer chemicals for a sustainable future.
