‘Forever chemicals’: Chinese scientists use AI to help detect unreported carcinogenic pollution around the world

Potentially carcinogenic chemicals are prevalent worldwide and going unreported, according to Chinese scientists who identified unknown “forever chemicals” in the environment with the help of machine learning.

The team’s new identification platform will make it possible to find and examine the impact of the synthetic chemicals in the environment, opening up opportunities for better risk management.

Many synthetic perfluoroalkyl and polyfluoroalkyl substances (PFAS) – which were widely used due to their resistance to oil, water and heat – have been restricted or banned around the world because of their links to organ damage, developmental issues and cancer.

“Banning legacy PFAS, however, has led to the increased use of PFAS substitutes that have not been adequately tested for safety, or publicly disclosed due to trade secret protection,” the researchers wrote in a paper published in the peer-reviewed journal Science Advances on Thursday.

“New PFAS are kind of like cousins to the old ones – they share similar properties and similar structural features, but [are] not completely consistent,” said Wei Si, the paper’s corresponding author and a professor at Nanjing University.

“Because they’re different, they’re not as familiar to scientists and regulators, which makes it more difficult to keep track of them and evaluate their potential risks,” Wei said, according to a release by the American Association for the Advancement of Science (AAAS) which publishes the journal.

Existing methods cannot identify many of the unknown PFAS chemicals, and there are no platforms capable of identifying unknown PFAS in environmental samples, according to the paper.

The team outlined how their innovation could help track them.

“Here, we report a PFAS identification platform that can discover and accurately identify unknown [PFAS] in environmental samples,” the team wrote.

The non-biodegradable “forever chemicals” – named for their persistence in the environment – became widely used around the world in all kinds of products – including non-stick pans, food packaging, and waterproof and fireproof clothing.

To screen for unknown PFAS, the team used a molecular network – a screening tool that analyses similarities in mass spectra, or plots of molecules and chemicals that reveal their signature and structure – that was enhanced using machine learning.

“By applying molecular networking, we can map out the relationships between known and unknown PFAS,” to identify similar structures, Wei told AAAS.

When their identification platform – called APP-ID – was used to examine samples of waste water from an industrial estate in Jiangsu province in eastern China, they discovered 733 PFAS belonging to 31 different classes of PFAS, according to the paper. Some 17 of the classes were previously unknown.

The team also used APP-ID to screen a global repository of environmental samples called MassIVE. By studying samples from 20 countries they discovered 126 PFAS, including 37 that had only recently been identified and including 81 unreported PFAS.

Their platform identified unknown PFAS with an accuracy of 58.3 per cent, compared to other platforms they compared to which had an accuracy of 12-44 per cent, according to AAAS.

Their method also had a false positive rate of 0.7 per cent – lower than other methods which ranged from 2.4 per cent to 46 per cent, the paper said.

“Our results show that APP-ID is effective and suggest that unknown PFAS contamination is as widespread as legacy PFAS,” the team wrote.

Their work “reveals a greater presence of these chemicals in the environment than previously known”, Wei told AAAS, adding that their finding would be “critical” for examining the impact of PFAS on the planet.

“APP-ID, which can reveal more unknown classes of PFAS in human and environmental exposome than current platforms, is expected to support a more precise chemical management and risk assessment of forever chemicals,” the team said.