6 minute read

Mitigating the environmental harm of PFAS ‘forever chemicals’

A new NSF-funded collaboration is focused on developing tools to identify products with PFAS, accelerate the search for PFAS substitutes with the help of generative AI, and recommend materials for capturing PFAS at manufacturing sites.

A new NSF-funded collaboration is focused on developing tools to identify products with PFAS, accelerate the search for PFAS substitutes with the help of generative AI, and recommend materials for capturing PFAS at manufacturing sites.

PFAS may not yet be a household word, but they are woven into the fabric of our lives. Prized for their water and grease-repelling properties, these synthetic chemicals can be found in nonstick pans, outdoor gear, food packaging, furniture, and even dental floss. Their ability to withstand harsh chemicals and punishing heat have also made them essential to making everything from computer chips to rechargeable batteries.

But PFAS, short for per- and polyfluoroalkyl substances, have also become pervasive. They’ve been found in our bodies, and in the water and soil of most places on Earth. Some can take thousands of years to break down in nature, and even small exposures over long periods can be harmful. PFAS have been linked to cancer in humans and animals, thyroid disease, hormonal dysfunction, and high cholesterol, among other problems.

In the search for solutions, the U.S. National Science Foundation is now funding an ambitious $5 million PFAS-replacement project called PFACTS, under its Convergence Accelerator initiative. IBM Research will lead a group of experts in academia, industry, and the non-profit sector in a multi-pronged effort to reduce the environmental impact of PFAS.

The project has three objectives: to simplify the identification of PFAS chemicals and materials; to speed up the discovery of safer PFAS alternatives with the help of generative AI; and to recommend methods for capturing PFAS from industrial applications until viable replacements can be found.

Nonstick, waterproof — and virtually indestructible

The first PFAS were introduced to the world in the 1940s. Their unique properties have led to their widespread use in everything from Teflon cookware to stain-resistant fabric, fire-fighting foam, medical devices, consumer electronics, semiconductors, and more.

The secret to their superpowers is an element in the upper right corner of the periodic table: fluorine. “It has unique properties that can’t be simply or easily replicated by any other element,” said Dan Sanders, a materials scientist and manager at IBM Research.

The carbon-fluorine bond is one of the strongest known to chemistry. “This allows PFAS molecules to withstand harsh chemicals, radiation, and the high heat and pressure involved in some industrial processes,” said Sanders. “But it also prevents these chemicals from breaking down in the environment or the human body.”

When they seep into the environment, PFAS latch onto soil particles and groundwater. Under President Joe Biden, the US Environmental Protection Agency (EPA) has expanded its monitoring and regulation of PFAS, and this month, passed a rule requiring municipal water systems to reduce to near-zero six of the most common PFAS from drinking water supplies.

PFAS get into our bodies via contaminated food and drinking water. But unlike other organic compounds that accumulate in fatty tissue, PFAS are drawn to proteins, leading to accumulations in the blood and liver. Structurally, they resemble fatty acids, which means they end up interacting with a lot of the same systems in the body, said project collaborator Carla Ng, a chemical engineer at the University of Pittsburgh. “That allows them to muck things up because they can’t actually be metabolized.”

Navigating the PFAS maze

Government agencies have sought for decades to regulate PFAS, even as the number of compounds have proliferated wildly. Under the most expansive definition, there are now more than 7 million reported varieties, with detailed toxicology information existing for only about a dozen.

Globally, PFAS are governed by a web of overlapping regulations and varying definitions, making it hard for companies to know which substances are in their products and supply chains, what rules apply, and how they should respond.

In the first phase of the NSF project, IBM and collaborator Digital Science, a London-based tech company, will gather into a single database all PFAS that have been reported in the scientific literature, along with their regulatory classifications. IBM researchers will also train large language models to comb US patent filings for PFAS, to match reported molecules with their real-world applications.

The database is meant to be a common reference point for regulators, researchers, and industry, and to give companies, especially, a clearer path through a labyrinth of policies. To make the database accessible, IBM will work with collaborator ChemForward, a nonprofit focused on safer chemistry adoption, to develop a user interface that companies can use to screen chemicals and materials in their supply chain for PFAS.

“A company could type in a bill of materials coming in their door and instantly get clarity on what PFAS are present and what rules apply,” said Stacy Glass, who heads and co-founded ChemForward.

As the database gets built out, companies will also be able to search for chemicals that can serve as PFAS substitutes as well as PFAS capture materials.

Generative AI and the search for safe PFAS substitutes

Finding PFAS replacements at an industrial scale is no easy task. Generative AI has potential to speed up the search if two main challenges can be overcome. One is predicting the health and environmental risks of replacement candidates. The other is engineering molecules that can match PFAS performance.

IBM’s foundation models could help on both counts. Researchers recently built into IBM’s MoLFormers family of models the capacity to consider molecular structure and physiochemical properties like polarity and solubility. This makes it easier to predict the toxicity and biodegradability of candidate molecules, and weed out those that may be hazardous or too long-lived.

IBM has started to use these tools to design safer replacements for the PFAS superacids used in chip photolithography. Researchers recently used an AI-driven framework to design 6,000 molecules structurally similar to superacids, but with less predicted toxicity. Currently, work is focused on expanding the model to consider various performance requirements and additional hazard dimensions.

“It’s a multi-dimensional optimization problem beyond a scale that humans can consider,” said Jed Pitera, a sustainable materials researcher at IBM. “Our hope is that by building large foundation models trained on lots of different data types, we can generate candidate molecules that are safer without any loss in performance.”

In the 2000s, the semiconductor industry, including IBM, replaced long-chain PFAS in their chip making processes for short chain PFAS which were thought at the time to be better for the environment. But recent evidence suggests that these compounds may also be hazardous.

To avoid a repeat scenario, researchers are trying to learn more about the mechanisms that make PFAS dangerous. Ng and her students are currently running simulations of how PFAS interact with proteins in the body. The simulation data will help steer generative AI models away from molecules with similar structures and disruptive effects. “Our mantra is to avoid regrettable substitutes,” she said.

Trapping PFAS at their source

With PFAS regulations tightening, and the search for replacements still in its early stages, it may be years before some companies can fully phase out PFAS from their operations. The third component of PFACTS is focused on identifying the best materials to remove these chemicals from industrial waste streams and prevent their release into the environment.

Researchers will evaluate many different options.

Two of the most common materials for removing PFAS from water can also be found in household water filters: activated carbon and synthetic ion-exchange resins. Activated carbon, which can be harvested from coconut shells, is typically used to remove chlorine, among other contaminants, while ion-exchange resins, eliminate minerals and other charged molecules.

Damian Helbling, a water quality engineer at Cornell University, will run simulations with his students to predict how readily various PFAS bind with activated carbon, ion exchange resins, and other novel capture materials. Their results will be incorporated into PFACTS to guide companies looking for removal solutions.

Fluorinated gases represent another potentially more urgent challenge. In addition to being a forever chemical, many PFAS gases are thousands of times more potent than carbon dioxide at heating up the atmosphere.

The properties that make fluorinated gases persist in the atmosphere also make them difficult to capture and destroy. “Imagine a gas resembling your Teflon-lined kitchenware — nothing sticks to it,” said project collaborator Ash Wright, a researcher with Numat Technologies in Chicago. “This is partly due to the molecule’s low polarizability, so it interacts poorly with other molecules and surfaces.”

Numat specializes in developing porous materials known as metal-organic frameworks (MOF) for capturing fluorinated gases. As part of PFACTS, Numat will run simulations to identify the best MOF variants to target gases like CF4 that lack the “stickiness” of other gases.

Separately, IBM will also apply generative AI toward the search for replacements for PFAS-based process gases.

Tangible solutions

The work of developing chemical substitutes typically takes a decade or more. With thousands of varieties in wide use, and unusual properties to replicate, PFAS could take longer. But the world has no time to wait.

“We need to do something immediately,” said Ng. “This is the time to make changes.”

Researchers are hopeful that the project, with its mix of perspectives from academia and industry, can lead to high impact discoveries. “The product isn’t necessarily academic papers,” said Helbling. “We’re working toward tangible solutions that could benefit society in important ways, relatively quickly.”