In the past few months, Baker’s team has been collaborating with biologists who have been trying to figure out the shape of the protein they are studying. “There are a lot of really cool biological studies that have really accelerated,” he said. A public database containing hundreds of thousands of ready-made protein shapes should be a bigger accelerator.
“It looks impressive,” said Tom Ellis, a synthetic biologist at Imperial College London who studies yeast genomes, and he is happy to try the database. But he warned that most of the predicted shapes have not yet been verified in the laboratory.
In the new version of AlphaFold, predictions come with a confidence score, which the tool uses to mark how close it thinks each predicted shape is to the real thing. Using this method, DeepMind found that AlphaFold predicts the shape of 36% of human proteins with accuracy to the level of a single atom. Hassabis said this is enough for drug development.
Previously, after decades of research, only 17% of the protein in the human body had its structure determined in the laboratory. If AlphaFold’s prediction is as accurate as DeepMind said, then the tool has more than doubled this number in just a few weeks.
Even predictions that are not completely accurate at the atomic level are still useful. For more than half of the protein in the human body, AlphaFold has predicted a shape that should be enough for researchers to figure out the function of the protein. The rest of AlphaFold’s current predictions are either incorrect or are for one-third of the proteins in the human body, which have no structure at all before they are combined with other proteins. “They are very soft,” Hassabis said.
Mohammed AlQuraish, a systems biologist at Columbia University, who has developed his own protein structure prediction software, said: “It can be applied at this level of quality. This fact is impressive.” He also pointed out that having a large biological body The structure of most proteins will make it possible to study these proteins as a system rather than working in isolation. “This is what I think is most exciting,” he said.
DeepMind is releasing its tools and forecasts for free and will not say whether there are plans to profit from them in the future. But this possibility is not ruled out. In order to establish and operate the database, DeepMind is cooperating with the European Molecular Biology Laboratory, an international research organization that already has a large protein information database.
For now, AlQuraishi can’t wait to see how the researchers deal with the new data. “It’s magnificent,” he said, “I don’t think any of us thought we would be here so soon. It’s really puzzling.”