The optimal artificial intelligence (AI) strategy for playing board games like Chess and Go is reinforcement learning. This approach has recently been adapted for a powerful protein design program.
By using this method, researchers have demonstrated that when it is trained to solve long-standing puzzles in protein science, software can excel at creating useful molecules. For example, in one experiment, proteins made with the new approach were found to be more effective at generating useful antibodies in mice compared with previous methods.
Reinforcement learning is a type of machine learning in which a computer program learns to make decisions by trying different actions and receiving feedback. By devising a suitable algorithm, such as one that can learn to play chess, the same process of assessing millions of different moves can be applied to biological experiments.
This approach has led scientists from University of Washington School of Medicine/UW Medicine to speculate that if this method is applied to the right research problems, then it likely could accelerate progress in a variety of scientific fields.
To explore the potential with protein design scientists gave a computer millions of simple starting molecules. The software proceeded to make ten thousand attempts at randomly improving each toward a predefined goal. The computer lengthened the proteins or bent them in specific ways until it learned how to contort them into desired shapes.
From this, the scientists manufactured hundreds of AI-designed proteins in the laboratory. By using electron microscopes and other instruments, the researchers confirmed that many of the protein shapes created by the computer were indeed realized in the laboratory. Each was confirmed to be of an atomically accurate design.
According to lead researcher Isaac D. Lutz, Shunzhi Wang: “Our approach is unique because we use reinforcement learning to solve the problem of creating protein shapes that fit together like pieces of a puzzle. This simply was not possible using prior approaches and has the potential to transform the types of molecules we can build.”
It is hoped that the new breakthrough will lead to more potent vaccines and possibly a new era in protein design. This includes running the spectrum from developing more effective cancer treatments to creating new biodegradable textiles.
It is anticipated that more accurate the technology becomes, then the more it opens up potential applications, including vascular treatments for diabetes, brain injuries, strokes, and other cases where blood vessels are at risk.
The research describing how scientists have successfully applied reinforcement learning to a challenge in molecular biology appears in the journal Science. The paper is titled “Top-down design of protein architectures with reinforcement learning.”
