The way in which predictions were made before the Omicron spike protein experiment reflects the recent dramatic changes in molecular biology brought about by artificial intelligence.The first software capable of accurately predicting protein structure was widely used months before Omicron appeared, thanks to Competition Research Team At Alphabet’s AI Lab in the UK Deep thinking And the University of Washington.
Ford used these two software packages, but because they were not designed or validated to predict small changes caused by mutations like Omicron, his results were more instructive than deterministic. Some researchers are skeptical of them. But the fact that he can easily experiment with powerful protein prediction artificial intelligence illustrates how recent breakthroughs have changed the way biologists work and think.
Subramaniam said that he received four or five emails from people who provided predicted Omicron spike structures while studying laboratory results. “A lot of people do this just for fun,” he said. Subramaniam said that direct measurement of protein structure will still be the ultimate measure, but he expects that artificial intelligence prediction will increasingly become the core of research—including research on future disease outbreaks. “This is transformative,” he said.
Because the shape of a protein determines its behavior, understanding its structure can help a variety of biological research, from evolutionary research to disease research. In drug research, clarifying the structure of proteins can help reveal potential targets for new therapies.
Determining the structure of a protein is far from simple. They are complex molecules assembled from instructions encoded in the organism’s genome, and can be used as enzymes, antibodies, and many other life machines. Proteins are made up of strings of molecules called amino acids, which can be folded into complex shapes that behave in different ways.
Deciphering the structure of a protein has traditionally involved hard laboratory work. Most of the approximately 200,000 known structures were drawn through a tricky process in which proteins form crystals and bombarded with X-rays. Newer technologies such as electron microscopes used by Subramaniam can be faster, but the process is still not easy.
At the end of 2020, after decades of slow progress, the long-term hope that computers can predict protein structure based on amino acid sequences suddenly became a reality. The DeepMind software called AlphaFold proved to be very accurate in the protein prediction competition, so much so that the co-founder of the challenge, University of Maryland professor John Moult, announced that the problem had been solved. “I have personally studied this issue for so long,” Moult said, and DeepMind’s achievement is “a very special moment.”
This moment also frustrated some scientists: DeepMind did not immediately release the details of how AlphaFold works. “You are in this strange situation, and you have made significant progress in your field, but you can’t build on it,” David Baker, who works on protein structure prediction in a laboratory at the University of Washington, Told WIRED last year. His research team used the clues provided by DeepMind to guide the design of an open source software named RoseTTAFold, which was released in June and is similar to AlphaFold but not as good as AlphaFold. Both are based on machine learning algorithms to predict protein structure by training a set of more than 100,000 known structures. Next month, DeepMind Announce details And released AlphaFold for anyone to use. Suddenly, the world has two methods of predicting protein structure.
Minkyung Baek, a postdoctoral researcher in Baker’s lab who led RoseTTAFold’s research, said that she was surprised how quickly protein structure prediction became the standard in biological research. Google Scholar reports that UW and DeepMind’s papers on their software were cited in more than 1,200 academic articles within a short period of time after publication.
Although it has not been proven that predictions are essential for Covid-19 research, she believes that they will become more and more important for dealing with future diseases. The answer to eliminating the pandemic will not be formed entirely by algorithms, but the structure of predictions can help scientists develop strategies. “The predictive structure can help you focus your experimental work on the most important issues,” Bai said. She is now trying to get RoseTTAFold to accurately predict the structure of antibodies and invading proteins when combined, which will make the software more useful for infectious disease projects.
Despite their impressive performance, protein predictors cannot reveal all the information about molecules. They spit out a single static structure for the protein, and did not capture the bending and swinging that occurs when it interacts with other molecules. These algorithms are trained on databases with known structures, which are more able to reflect those databases that are easiest to draw through experiments, rather than the full diversity of nature. Kresten Lindorff-Larsen, a professor at the University of Copenhagen, predicts that these algorithms will be used more frequently and will be useful, but he said: “When these methods fail, we as a field also need to learn better.”