Some scientists spend their lives determining the shape of tiny proteins in the human body.
Proteins are the microscopic mechanisms that control the behavior of viruses, bacteria, the human body and all living things. They start out as chains of chemical compounds before twisting and folding into three-dimensional shapes that define what they can and cannot do.
For biologists, identifying the exact shape of a protein often requires months, years, or even decades of experimentation. It takes skill, intelligence, and more than a little elbow fat. Sometimes they never succeed.
Now an artificial intelligence laboratory in London has built a computer system that does the job in hours – maybe even minutes.
DeepMind, a laboratory owned by the same parent company as Google, announced on Monday that its system called AlphaFold had solved the so-called “protein folding problem”. Given the range of amino acids that make up a protein, the system can predict its three-dimensional shape quickly and reliably.
This long-awaited breakthrough could accelerate the ability to understand diseases, develop new drugs, and unravel the mysteries of the human body.
Computer scientists have had problems building such a system for more than 50 years. For the past 25 years, they have measured and compared their efforts against a global competition called the Critical Assessment of Structure Prediction (CASP). So far, no candidate had come close to solving the problem.
DeepMind solved the problem with a variety of proteins and achieved a level of accuracy that rivaled physical experimentation. Many scientists had assumed that the moment was years, if not decades, away.
“I’ve always hoped I would live to see that day,” said John Moult, a professor at the University of Maryland who helped found CASP in 1994 and continues to oversee the biennial competition. “But it wasn’t always clear that I would make it.”
As part of this year’s CASP, the DeepMind technology was presented by Dr. Moult and other researchers overseeing the competition.
If DeepMind’s methods can be refined, they and other researchers could accelerate the development of new drugs as well as efforts to apply existing drugs to new viruses and diseases.
The breakthrough comes too late to have a significant impact on the coronavirus. However, researchers believe that DeepMind’s methods could speed up the response to future pandemics. Some believe it could also help scientists better understand genetic diseases such as Alzheimer’s or cystic fibrosis.
However, experts warned that this technology would affect only a small part of the long process by which scientists identify new drugs and analyze diseases. It was also unclear when or how DeepMind would share its technology with other researchers.
DeepMind is a key player in a profound change that has spread across science, technology, and the medical community over the past 10 years. Thanks to an artificial intelligence known as a neural network, machines can now learn to perform many tasks that used to be out of their reach – and sometimes out of the reach of humans.
A neural network is a mathematical system loosely modeled on the network of neurons in the human brain. It learns skills by analyzing large amounts of data. For example, by finding patterns in thousands of cat photos, one can learn to recognize a cat.
This is the technology that recognizes faces in the photos you post on Facebook, identifies the commands you bark into your smartphone, and translates one language to another via Skype and other services. DeepMind uses this technology to predict the shape of proteins.
If scientists can predict the shape of a protein in the human body, they can determine how other molecules bind or physically bind to it. This is one way drugs are developed: a drug binds to certain proteins in your body and changes their behavior.
By analyzing thousands of known proteins and their physical shapes, a neural network can learn to predict the shapes of others. With this method, DeepMind took part in the CASP competition for the first time in 2018 and its system outperformed all other competitors, which signaled a significant change. However, the team of biologists, physicists, and computer scientists led by a researcher named John Jumper were far from able to solve the ultimate problem.
In the two years since then, Dr. Jumper and his team developed a completely new neural network specifically for protein folding, which resulted in an enormous leap in accuracy. Their latest version offers a powerful, albeit incomplete, solution to the problem of protein folding, said DeepMind researcher Kathryn Tunyasuvunakool.
The system can accurately predict the shape of a protein about two-thirds of the time according to the results of the CASP competition. And his errors with these proteins are smaller than the width of an atom – an error rate that can keep up with physical experiments.
“Most of the atoms are within an atomic diameter that they are in the experimental structure,” said Dr. Moult, the organizer of the competition. “And with those who are not, there are other possible explanations for the differences.”
Andrei Lupas, Director of the Department of Protein Development at the Max Planck Institute for Developmental Biology in Germany, is one of AlphaFold’s employees. He’s part of a team that spent a decade trying to determine the physical form of a particular protein in a tiny bacteria-like organism called the archaeon.
This protein spans the membrane of individual cells – part is inside the cell, part is outside – and that’s what makes it to scientists like Dr. Lupas find it difficult to determine the form of the protein in the laboratory. Even after a decade, he couldn’t pinpoint the shape.
With AlphaFold he solved the problem in half an hour.
If these methods continue to improve, they could be a particularly useful way to determine whether a new virus can be treated with a cocktail of existing drugs.
“We could start by studying any compound that is approved for use in humans,” said Dr. Lupas. “We could face the next pandemic with the drugs we already have.”
During the current pandemic, a simpler form of artificial intelligence has proven helpful in some cases. A system developed by another London company, BenevolentAI, helped identify an existing drug, baricitinib, that could be used to treat seriously ill Covid-19 patients. The researchers have now completed a clinical study, but the results have not yet been published.
As researchers keep improving the technology, AlphaFold could further accelerate this type of drug reuse, as well as the development of entirely new vaccines, especially if we come across a virus even less well known than Covid-19.
David Baker, the director of the Institute of Protein Design at the University of Washington, who has used similar computer technologies to develop anti-coronavirus drugs, said DeepMind’s methods could expedite that work.
“We were able to develop coronavirus-neutralizing proteins in several months,” he said. “But our goal is to do something like this in a couple of weeks.”
Still, the pace of development has to deal with other issues like massive clinical trials, said Dr. Vincent Marconi, a researcher at Emory University in Atlanta who led the baricitinib study. “It takes time,” he said.
However, DeepMind’s methods can, at least in some cases, determine whether a clinical trial is failing due to toxic reactions or other issues.
Demis Hassabis, executive director and co-founder of DeepMind, said the company had planned to release details about its work but that was unlikely by next year. He also said the company is exploring ways to share the technology with other scientists on its own.
DeepMind is a research laboratory. No products are sold directly to other laboratories or companies. However, it could work with other companies to share access to its technology over the internet.
The lab’s biggest breakthroughs in the past have involved games. Systems were developed that surpassed the human performance of the old strategy game Go and the popular video game StarCraft – enormous technical achievements with no practical application. Now the DeepMind team is eager to bring its artificial intelligence technology to the real world.
“We don’t want to be a ranking company,” said Dr. Jumper. “We want real biological relevance.”