The way the predictions run ahead of the Omicron mutant protein experiments reflects a recent shift in molecular biology brought about by AI. The first software capable of accurately predicting protein structure was widely available just a few months before Omicron appeared, thanks to competitive study groups at Alphabet’s UK-based AI lab DeepMind and at the University of Washington.
Ford used both packages, but since neither has been designed or validated to predict small mutation-induced changes like Omicron’s, his results are more suggestive than definitive. definitively. Some researchers have treated them with suspicion. But the fact that he can easily experiment with powerful protein-predicting AI illustrates how recent breakthroughs are changing the way biologists work and think.
Subramaniam said he received four or five emails from proponents of the predicted Omicron spike structures while working on the results in his lab. “A lot of people did this just for fun,” he said. Subramaniam says direct measurements of protein structure will remain the ultimate measure, but he expects AI predictions to become increasingly central to research — including disease outbreaks in the future. future. “It can be transformative,” he said.
Because the shape of a protein determines how it functions, knowing its structure can help with all kinds of biological research, from the study of evolution to the study of disease. In drug research, finding protein structures can help reveal potential targets for new treatments.
Determining the structure of a protein is far from simple. They are complex molecules assembled from instructions encoded in an organism’s genome to act as enzymes, antibodies and many other machinery of life. Proteins are made of chains of molecules called amino acids that can be folded into complex shapes that work in different ways.
Deciphering the structure of a protein has traditionally involved painstaking laboratory work. Most of the approximately 200,000 known structures have been mapped by a complex process in which proteins are formed into a crystal and bombarded with X-rays. Newer techniques such as microscopy The electronics used by Subramaniam may be faster, but the process is still far from easy.
At the end of 2020, the long-held hope that computers could predict the structure of a protein from a sequence of amino acids suddenly became a reality, after decades of slow progress. The DeepMind software called AlphaFold has proven so accurate in a protein prediction contest that the challenge’s co-founder John Moult, a professor at the University of Maryland, claims the problem has been solved. . “Having worked personally on this issue for a long time,” says Moult, DeepMind’s achievement is “a very special moment.”
The timing also disappointed some scientists: DeepMind didn’t immediately release details on how AlphaFold works. “You’re in a strange situation where there’s this big leap in your field, but you can’t build on it,” said David Baker, whose lab at the University of Washington works on prediction. protein structure, told WIRED last year. His research team used clues given by DeepMind to guide the design of open source software called RoseTTAFold, which was released in June, which is similar but not as powerful as AlphaFold. Both are based on machine learning algorithms honed to predict protein structures by training on a collection of over 100,000 known structures. Next month, DeepMind Details are published own and release AlphaFold for everyone to use. Suddenly, the world had two ways to predict protein structure.
Minkyung Baek, a postdoctoral researcher at Baker’s lab who led the RoseTTAFold study, said she was surprised by the protein structure predictions that have quickly become the norm in research. biological rescue. Google Scholar reports that UW and DeepMind papers on their software have together been cited by more than 1,200 academic papers in the short time since they appeared.
While the predictions have not been shown to be important for Covid-19, she believes they will become increasingly important for future disease response. Pandemic answers won’t be made entirely of algorithms, but predicted structures can help scientists strategize. “A predictable structure can help you put your testing efforts on the issues that matter most,” says Baek. She is now trying to get RoseTTAFold to accurately predict the structure of antibodies and the invading proteins when bound together, which will make the software more useful for infectious disease projects.
Despite their impressive performance, protein prediction tools don’t reveal everything about a molecule. They created a unique static structure for a protein, and failed to capture the bending and wobbling that take place when it interacts with other molecules. Algorithms are trained on databases of known structures, more reflecting structures that are easier to map experimentally than nature’s full diversity. Kresten Lindorff-Larsen, a professor at the University of Copenhagen, predicts algorithms will be used more often and will be useful, but said, “We as a field need to learn better as methods This failed.”
