Researchers behind Google DeepMind’s AlphaFold program have landed one of the most prestigious prizes in science for solving a grand challenge in biology that stood for half a century.
Demis Hassabis and John Jumper, who led the development of AlphaFold, an artificial intelligence program, share the $250,000 Lasker basic medical research award for their “revolutionary technology” to predict the 3D shapes of proteins.
Beyond recognising the substantial impact of AlphaFold on modern science, the award is noteworthy because many Lasker winners go on to receive the Nobel prize, raising the prospect of AI research earning a Nobel for the first time. In the past 20 years, 32 Lasker winners have received a Nobel.
“It’s humbling to be recognised by a prize with such a distinguished history, and it’s not just Demis and me, it’s the whole team,” Jumper told the Guardian. “One of the things that’s really significant about this absolutely major award is that it’s for medicine.”
As arcane as the problem sounds, the importance of protein folding is hard to overstate. Proteins are the building blocks of life and the shapes they form underpin their function, from how antibodies fight viruses to how insulin controls blood sugar levels. Working out the structures in a lab can involve years of painstaking experiments, but AlphaFold reached its answers in minutes.
Last year, the AlphaFold team released more than 200m predicted structures, amounting to nearly all the proteins known to science. The database has turbocharged research across the sciences, from finding drugs for neglected diseases, to developing a “molecular syringe” to inject human cells, and understanding the horrendously complex nuclear pore, which controls all that enters and leaves a cell nucleus.
AlphaFold came hot on the heels of AlphaGo, the AI that beat Lee Sedol, one of the world’s leading players of the Chinese strategy game Go.
AlphaFold learned how to predict protein structures from their chemical makeup by training on 170,000 protein sequences and the structures scientists had worked out in the lab.
The first test of AlphaFold came in 2018 when the team entered the program for a biennial “protein Olympics” known as Casp, the Critical Assessment of Protein Structure Prediction. The winning entrants are those best able to predict protein structures from their constituent amino acids.
While AlphaFold fared well at the challenge, it was not good enough for scientists to use. Work to improve the AI hit a brick wall, but eventually the research team ramped up its performance and in 2020 the program won the Casp competition.
“I don’t think we knew, going in, how hard it would be, or how much it would come not from a silver bullet exactly but from lots of ideas,” Jumper said. AlphaFold’s impact on research was beyond his wildest dreams, he added.
Prof John Moult, a computational biologist at the University of Maryland and chair of Casp, said AlphaFold was “instantly and obviously a very big deal”.
“It’s hard to exaggerate the impact,” he said. “I don’t know any structural biologists who are not using it, one way or another. And the impact is not just in structural biology.”
Asked where the research would go next, Jumper said it was crucial to focus on the right challenge. “There are many, many more problems. Which are going to produce this AlphaFold-like avalanche of downstream work done by others that’s really important?” he said. “We need to find the right ones and have that impact. This is by no means the end of science, or the end of biology.”
