AI has cracked a problem that stumped biologists for 50 years. It’s a huge deal.

DeepMind, an AI research lab that was bought by Google and is now part of the parent company of Google Alphabet, this week announced a major breakthrough that an evolutionary biologist called a “changer of game ”.

“It will change medicine,” said biologist Andrei Lupas. Nature. “It will change research. It will change bioengineering. It will change everything.

Discovery: DeepMind claims that its AI system, AlphaFold, has solved “the protein folding problem” – a big challenge in biology that has annoyed scientists for 50 years.

Proteins are the basic machines that make your cells work. They start out as chains of amino acids (imagine the beads on a necklace), but they quickly fold back into a unique three-dimensional shape (imagine you rubbing the pearl necklace in your hand).

This 3D shape is crucial because it determines how the protein works. If you are a scientist developing a new drug, you want to know the shape of the protein, as this will help you find a molecule that can bind to it. inside to change its behavior. The problem is, it’s extremely difficult to predict what shape a protein will take.

Every two years, researchers working on this problem attempt to prove the quality of their predictive powers by submitting a prediction about the forms certain proteins will take. Their contributions are judged at the Critical Assessment of Structure Prediction (CASP) conference, which is essentially a sophisticated science competition for adults.

By 2018, DeepMind’s AI was already surpassing everyone at CASP, causing melancholy feelings among human researchers. DeepMind took the win that year, but that still hadn’t solved the protein folding problem. Not even close.

This year, however, his AlphaFold system was able to predict – with impressive speed and accuracy – what shapes amino acid chains would fold into. AI isn’t perfect, but it’s pretty awesome: when it makes mistakes, it’s usually only shifted by an atom’s width. This is comparable to the errors you get when doing physical experiments in a lab, except those experiments are much slower and much more expensive.

“It’s a big deal,” said John Moult, who co-founded and oversaw CASP. Nature. “In a sense, the problem is solved.”

Why this is a big problem for biology

AlphaFold technology has yet to be refined, but assuming researchers can pull it off, this breakthrough will likely accelerate and improve our ability to develop new drugs.

Let’s start with speed. To get an idea of ​​how much AlphaFold can speed up the work of scientists, consider the experience of Andrei Lupas, evolutionary biologist at the Max Planck Institute in Germany. He spent a decade – a decade! – try to understand the shape of a protein. But no matter what he tried in the lab, the answer eluded him. Then he tried AlphaFold and got the answer within half an hour.

AlphaFold has implications for everything from Alzheimer’s disease to future pandemics. This can help us understand diseases, as many (like Alzheimer’s disease) are caused by misfolded proteins. This can help us find new treatments and also quickly determine which existing drugs can be usefully applied, for example, to a new virus. When another pandemic strikes, it could be very helpful to have a system like AlphaFold in our back pocket.

“We could start screening for all compounds cleared for use in humans,” Lupas said. Told the New York Times. “We could face the next pandemic with the drugs we already have.”

But for that to be possible, DeepMind would have to share its technology with scientists. The laboratory said he’s exploring ways to do it.

Why this is a big deal for artificial intelligence

Over the past few years, DeepMind has made a name for itself playing games. He built AI systems that crushed pro gamers in strategy games like StarCraft and go. Much like the chess matches between IBM’s Deep Blue and Garry Kasparov, these matches have mainly served to prove that DeepMind can create an AI that surpasses human abilities.

Now DeepMind is proving he’s grown up. He went from video games to solving scientific problems of real significance – problems that can be life or death.

The protein folding problem was a perfect thing to solve. DeepMind is a world leader in creating neural networks, a type of artificial intelligence loosely inspired by neurons in a human brain. The beauty of this type of AI is that it doesn’t require you to pre-program it with a lot of rules. You just need to feed a neural network enough examples of something, and it can learn to detect patterns in the data and then draw inferences from them.

So, for example, you can present it with several thousand chains of amino acids and show it in what shape they folded. Gradually, it detects patterns of how given chains tend to form – patterns that human experts may not have detected. From there, he can make predictions of how the other chains will bend.

This is exactly the kind of problem neural networks excel at, and DeepMind has recognized this, by matching the right kind of AI with the right kind of puzzle. (He also incorporated more complex knowledge – on physics and evolutionary amino acid sequences, for example – although the details remain rare as DeepMind is still preparing a peer-reviewed article for publication.)

Other laboratories have already harnessed the power of neural networks to make breakthroughs in biology. Earlier this year, AI researchers formed a neural network by providing data on 2,335 molecules known to have antibacterial properties. Then they used it to predict which other molecules – out of 107 million possibilities – would have these properties as well. In this way, they succeeded in identifying new types of antibiotics.

DeepMind researchers close the year with another achievement that shows just how much AI has matured. This is really great news for a generally terrible 2020.

Subscribe to the Future Perfect newsletter and we’ll send you a roundup of ideas and solutions to tackle the world’s biggest challenges – and how to improve your job.