ProteoGen: The AutoML System That Designs Novel Enzymes to Eat Forever Chemicals in Weeks

The Indestructible Molecules

Imagine you build a toy out of the strongest, most unbreakable plastic in the world. You play with it for a while, and then you throw it away. But because it is so strong, it never breaks down. It sits in the garbage dump for a hundred years, then a thousand years, then a million years. It just sits there, taking up space, slowly leaking tiny, invisible pieces into the soil and the water. This is the problem of "forever chemicals," known scientifically as PFAS. We put them in our non-stick pans, our waterproof jackets, and our fire-fighting foam because they are incredibly good at repelling water and oil. But that same superpower makes them indestructible in nature. They do not rot, they do not degrade, and they accumulate in our blood, causing terrible diseases. For decades, scientists have tried to find a way to break them apart, but the chemical bonds are too strong. But in June 2026, an automated machine learning system called "ProteoGen" designed a brand-new, artificial enzyme that eats PFAS for breakfast, solving a crisis that humans created.

To understand how ProteoGen works, we have to think about enzymes as tiny, biological scissors. In nature, bacteria use enzymes to cut up food, breaking complex molecules into smaller pieces they can eat. But natural bacteria have never encountered PFAS before, so they do not have the right "scissors" to cut them. Human chemists have tried to build synthetic catalysts to break PFAS, but it requires massive amounts of energy, extreme heat, and toxic acids. It is like trying to cut a diamond with a sledgehammer. What we needed was a biological scissor that was perfectly shaped to cut the exact, specific chemical bond of a PFAS molecule, and we needed it to work at room temperature in a glass of water. Designing such a protein from scratch is so complex that it would take a team of human biochemists centuries to guess the right shape.

The AI That Folds Proteins in its Mind

ProteoGen is an "AutoML" (Automated Machine Learning) system specifically built for protein design. It uses a technique called "diffusion modeling," similar to the AI that generates images, but instead of generating pixels, it generates atoms. ProteoGen starts with a blank canvas. It is given a target: the exact 3D structure of a PFAS molecule, and the specific chemical bond that needs to be broken. The AI then "dreams" up millions of different protein structures. It folds and unfolds them in its digital mind, checking if the active site of the protein—the "mouth" of the scissor—perfectly matches the shape of the PFAS molecule.

But ProteoGen does not just guess; it simulates the physics of the interaction. It calculates the quantum mechanical forces between the atoms of the enzyme and the atoms of the PFAS. It checks if the enzyme can actually pull the bond apart. After iterating through billions of designs, refining the shape by tiny fractions of an angstrom, ProteoGen outputs a sequence of amino acids that has never existed in the history of life on Earth. It then sends this digital blueprint to an automated robotic lab. The robot synthesizes the DNA for this new protein, inserts it into a harmless strain of bacteria, and the bacteria start churning out the new enzyme. From the initial concept to a physical vial of the enzyme took only three weeks.

Cleaning Up the Planet's Mistakes

The enzyme, named "PFAS-Zero," is a miracle of biological engineering. When it is introduced to water contaminated with forever chemicals, it latches onto the PFAS molecules. With a precise, mechanical twist, it snaps the incredibly strong carbon-fluorine bond, which is the hardest bond to break in organic chemistry. The PFAS molecule falls apart into harmless, natural salts and fluoride ions that are easily filtered out. The enzyme itself is completely biodegradable and non-toxic. Environmental cleanup crews are now deploying massive floating bioreactors filled with PFAS-Zero into contaminated rivers and lakes. The water goes in dirty, full of indestructible chemicals, and comes out pure and clean on the other side.

The implications for human health are profound. We are finally able to undo the chemical pollution of the last fifty years. The military is using the enzyme to clean up old airbase firefighting foam sites, which were some of the worst contamination hotspots in the world. Municipal water treatment plants are integrating the enzyme into their filtration systems, ensuring that the water coming out of the tap is 100% free of forever chemicals. ProteoGen has proven that the same industrial mindset that created the pollution can be used to engineer our way out of it. We are no longer just passive victims of our chemical legacy; we are the active architects of a clean future, guided by the infinite creativity of machine learning.