AI can design extremely dense protein shells that could one day lead to more potent vaccines.
The genetic material of viruses is housed in protein shells. Similar shells made in the lab are used in vaccines, encapsulating molecules that induce an immune response in the body. The chemical and biological properties of these artificially made shells depend on their construction – any imperfections in them, no matter how small, make them less effective, causing them to be unstable and react unpredictably inside cells. Isaac Lutz at the University of Washington in Seattle and his colleagues wanted to see if using artificial intelligence could make the design and creation of these shells more precise.
They first fed the AI the properties they wanted a shell to have, like its size and porosity. The AI then used reinforcement learning – the same iterative process AI systems use to learn to play games like chess by trying different moves, then receiving feedback and trying again. Here, the AI’s moves were combining, folding and intertwining small protein structures called alpha helices into 20-sided shells, then checking whether the design had the desired properties.
After the AI designed hundreds of thousands of shells, the researchers created about 350 of them in the lab. They examined them with an electron microscope and found the AI had made more dense shells than had ever been created. Lutz says this is because it started with very small building blocks that could be made to fit together more neatly than bigger protein structures that researchers previously used.
“It’s like we previously had to buy something like protein Legos first, and what you could build was limited with how they could fit together. Whereas now we can say what we want to build, then the AI designs and connects the exact Legos needed to accomplish that,” he says.
To test how high density affects the shells’ uses in living cells, the team studded the shells with different molecules and inserted them into mice. Notably, in one experiment with molecules that trigger production of influenza antibodies, the AI-designed shells resulted in a small but statistically significant increase in immune response compared with some more conventional vaccine candidates that are currently in clinical trials. Lutz says that this is because of the precision of the AI method – every molecule is exactly where it needs to be on the shell and the shell is structurally sound enough to support many of them.
“It’s astounding that the team could do this. It takes evolution billions of years to design single proteins that fold just right, but this is another level of complexity, to fold proteins to fit so well together and make closed structures,” says Martin Noble at Newcastle University in the UK.
Yang Zhang at the University of Michigan says that in addition to vaccines, AI-designed protein shells could be useful for gene therapy where genetic material could be placed inside of a shell tailored so that the patient’s cells don’t react to it adversely.