AI can design new proteins that unlock new healing materials
The new tool, ProteinMPNN, was described by a team of researchers from the University of Washington in two papers published in the journal Science. today (available here and here), provides a powerful addition to that technology.
The papers are the latest example of how deep learning is revolutionizing protein design by providing scientists with new research tools. Traditionally, researchers design proteins by tweaking those that occur naturally, but ProteinMPNN opens up a whole new universe of possible proteins for researchers to design from scratch.
“In nature, proteins solve basically all of life’s problems, from capturing energy from sunlight to making molecules. David Baker, one of the scientists behind the paper and director of the Institute for Protein Design at the University of Washington, says everything in biology happens from proteins.
“They evolve in evolution to solve problems that organisms face in evolution. But we face new problems today, like covid. If we can design proteins that are as good at solving new problems as proteins that evolved in evolution are to solve old problems, it would be really powerful.”
Proteins consist of hundreds of thousands of amino acids linked together into long chains, which then fold into a three-dimensional shape. AlphaFold helps researchers predict the outcome structure, providing insight into how they will behave.
ProteinMPNN will help researchers solve the inverse problem. If they had the correct protein structure in mind, it would help them figure out the amino acid sequence that folds into that shape. The system uses a neural network trained on a large number of amino acid sequence examples, which fold into a three-dimensional structure.
But researchers also need to tackle another problem. To design proteins that solve real-world problems, such as a new enzyme that digests plastic, they must first figure out which protein’s backbone will have that function.
To do that, researchers in Baker’s lab use two machine learning approaches, detailed in a article In the journal Science last July, the team called it “limited illusions” and “in painting.”