Toxic chemicals can be detected with new AI method

Swedish researchers from Chalmers University of Technology and the University of Gothenburg have devised a novel AI method adept at detecting toxic chemicals based solely on their molecular structures. This innovation not only promises to enhance our grasp on the myriad of chemicals pervading society but also offers a pathway to curtail animal testing.

Chemical usage is ubiquitous in our daily lives, spanning from household items to industrial processes, with repercussions extending to our water bodies and ecosystems. PFAS, for instance, a group of concerning substances, has been detected in alarming concentrations in groundwater and drinking water, posing significant risks due to its prevalent use in various products.

Despite stringent chemical regulations, adverse effects on humans and the environment persist, necessitating time-consuming animal testing for safety assessments. In the European Union alone, over two million animals undergo testing annually to comply with regulatory standards. Moreover, the rapid pace of new chemical development compounds the challenge of identifying potentially harmful substances.

The Swedish researchers' innovative method harnesses artificial intelligence to swiftly and cost-effectively evaluate chemical toxicity, offering a means to identify hazardous substances early on and minimize reliance on animal testing. By analysing extensive datasets from past laboratory tests, the AI model has been trained to accurately predict toxicity for previously untested chemicals.

With over 100,000 chemicals in circulation, conventional testing methods, including animal trials, are impractical for assessing toxicity comprehensively. The researchers foresee their method as a viable alternative, particularly beneficial for environmental research and stakeholders involved in chemical development and regulation. By making the method publicly available, they aim to broaden its accessibility and utility.

While computational tools for identifying toxic chemicals exist, their limited applicability and accuracy hinder widespread adoption. In comparative studies, the new AI method exhibits superior accuracy and versatility, thanks to its foundation in advanced deep learning techniques.

The AI model's effectiveness lies in its utilization of transformers, originally designed for language processing, which have demonstrated remarkable proficiency in deciphering chemical structures. By discerning molecular properties indicative of toxicity, the model can predict a compound's harmful effects through a deep neural network.

Looking ahead, the researchers anticipate AI systems gradually supplanting traditional laboratory tests, leading to reduced animal experimentation and cost savings in chemical development. This shift not only expedites the screening of vast datasets but also facilitates the identification of safer alternatives to toxic substances, thereby mitigating the adverse impacts of chemical pollution on both human health and ecosystem integrity.

Published in Science Advances, this research signifies a transformative leap in chemical toxicity assessment, enabled by the convergence of artificial intelligence and molecular science. Led by Mikael Gustavsson and Erik Kristiansson, the interdisciplinary collaboration underscores the potential of AI-driven innovations to address pressing environmental and public health challenges.