Deepmind, an AI research lab that Google bought and is now an independent part of Google’s parent company Alphabet, announced a major breakthrough this week that a developer biologist called it a “game changer.”
“This drug will change,” said biologist Andre Lupes Nature. “It simply came to our notice then. It will change bioengineering. It will change everything. ”
Progress: Deepmind says its AI system, Alphafold, has solved the “protein folding problem” – a major biological challenge that Scientists have wasted scientists for 0 years.
Proteins are the basic machines that complete the work in your cells. They start out as a string of amino acids (imagine a necklace on a necklace) but they soon join into a unique three-dimensional shape (imagine cutting a necklace in your hand).
That 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 a protein because it will help you come up with a molecule to build on it, right It’s about changing his behavior. The difficulty is predicting what shape the protein will take is incredibly difficult.
Every two years, researchers working on this problem have tried to prove how good their predictive powers are by submitting predictions about specific protein-taking shapes. Their entries are critically evaluated at the Stats Forecast (CASP) conference, which is basically a fancy science contest for adults.
By 2018, Deepmind’s AIA was already doing well to everyone in CASP, provoking some melancholy feelings among human researchers. Deepmind won that year, but it still hasn’t solved the protein folding problem. Not even close.
This year, however, its alphafold system was able to predict – with impressive speed and accuracy – in what shape the strings of amino acids would participate. AI isn’t perfect, but it’s pretty cool: when it makes mistakes, it’s usually just closed by the width of the atom. It is comparable to the errors you find when doing physical experiments in the lab, except that those experiments are slower and more expensive.
“This is a big deal,” said John Moult, who praised and overseen the CASP. Nature. “In a sense, the problem is solved.”
Why this is a big deal for biology
Alphafold technology still needs refining, but researchers assume it can pull this off, this advancement will probably speed up and improve our ability to develop new drugs.
Let’s start with speed. To gain an understanding of how Alphafold accelerates the work of scientists, consider the experience of Andre Lupus, an evolutionary biologist at the Max Planck Institute in Germany. He spent a decade – a decade! – Trying to figure out the shape of a protein. But no matter what he tried in the lab, the answer avoided him. Then he tried Alphafold and in half an hour he got the answer.
Alphafold has an impact on everything from Alzheimer’s disease to future epidemics. It can help us understand diseases, as many (such as Alzheimer’s) are caused by the wrong protein. It can help find new treatments and help us determine which existing drugs may be useful, for example, a new virus. When another epidemic strikes, having an alphafold-like system in our back pocket can be very helpful.
“We can start screening every compound that is licensed for use in humans,” Lupus told the New York Times. “We can already deal with the next epidemic with drugs.”
But for this to be possible, Deepmind will have to share it with its technical scientists. The lab says it is exploring ways to do that.
Why this is a big deal for artificial intelligence
Over the past few years, DeepMind has made a name for itself by playing sports. He has created AI systems that have crushed pro gamers in strategy games like Starcraft and Go. Like the chess match between IBM’s Deep Blue and Gary Kasparov, the match was largely about proving that Deepmind can create AI that transcends human capabilities.
Now, Deepmind is proving it has grown. He has graduated from playing video games to addressing scientific problems with real-world significance – problems that can be life-and-death.
Protein folding was an ideal thing to solve the problem. Deepmind is a world leader in creating neural networks, a type of artificial intelligence induced by the release of neurons in the human brain. The beauty of this type of AI is that you don’t have to preprogram it with too many rules. Simply feed a neural network some sufficient examples, and it can learn to find patterns in the data, then draw a list based on it.
So, for example, you can present it with thousands of strings of amino acids and show what shape they are connected to. Gradually, it finds patterns in the way it shapes a given wire – examples that human experts cannot detect. From there, it can predict how other strings will fold.
This is exactly the kind of problem in which neural networks excel, and Deepmind recognized that, with the right kind of AI. To marry, the right kind of puzzle. (It also consolidates some more complex knowledge – about amino acid sequences related to physics and evolution, for example – although details remain short as Deepmind is still preparing a peer-reviewed paper for publication.)
Other laboratories have used the power of neural networks to advance advances in biology. Earlier this year, AI researchers trained the neural network by providing information on 2,335 molecules with antibacterial properties. They then used it to predict which of the 107 million potential – which other molecules would also have these properties. In this way, they succeeded in identifying new types of antibiotics.
Deepmind researchers are linking the year to another achievement that shows how mature AI is. That’s really great news for the generally terrible 2020.
Sign up for Future Perfect Newsletter And we’ll send you a roundup of ideas and solutions to overcome the world’s biggest challenges – and how to get better.
