Home https://server7.kproxy.com/servlet/redirect.srv/sruj/smyrwpoii/p2/ Science https://server7.kproxy.com/servlet/redirect.srv/sruj/smyrwpoii/p2/ DeepMind solves the “big challenge” of folding proteins with AI AlphaFold

DeepMind solves the “big challenge” of folding proteins with AI AlphaFold



Demis Hasabis, CEO of Alphabet, a Google DeepMind research group, at the Google Future of Go Summit in China on May 23, 2017.

LONDON – Alphabet-owned DeepMind has developed artificial intelligence software that can accurately predict the structure in which proteins will fold in a matter of days, solving a 50-year “big challenge”

; that could pave the way for better understanding. of diseases and drug detection.

Every living cell has thousands of different proteins inside that keep it alive and healthy. Predicting the shape in which a protein will fold is important because it determines their function, and almost all diseases, including cancer and dementia, are related to how proteins function.

“Proteins are the most beautiful, magnificent structures, and the ability to predict exactly how they fold is really very, very challenging and has occupied many people for many years,” said Professor Dame Janet Thornton of the European Institute of Bioinformatics during a conversation. .

The British research laboratory DeepMind “AlphaFold” AI system participates in a competition organized by a group called CASP (Critical Assessment for Structure Prediction). This is a community experimental organization with a mission to accelerate the solution of one problem: how to calculate the 3D structure of protein molecules.

CASP, which has been monitoring progress in this area for 25 years, compares competition applications to an “experimental gold standard”. On Monday, she said DeepMind’s AlphaFold system had achieved unparalleled levels of accuracy in predicting protein structure.

“DeepMind has jumped ahead,” said Professor John Molt, chairman of CASP, during a press conference ahead of the announcement. “50 years of great challenge in computer science has been largely solved.”

Molt added that there are “big impacts a little further down the line in drug design” in the emerging field of protein design.

With about 1,000 employees and almost no revenue, DeepMind has become an expensive support company for Alphabet (Google’s parent). However, it has emerged as one of the leaders in the global artificial intelligence race, along with Facebook-like AI Research, Microsoft and OpenAI.

The breakthrough was welcomed by Google CEO Sundar Pichai on Twitter.

DeepMind co-founder and CEO Demis Hasabis said in a call: “The ultimate vision behind DeepMind has always been to build a common AI and then use it to help us better understand the world around us by significantly accelerating the pace. of scientific discoveries. ”

The company, which Google bought for $ 600 million in 2014, is best known for creating AI systems that can play games like Space Invaders and the ancient Chinese board game Go. However, he has always said that he wants to have a greater scientific impact.

“Games are a great testing ground for effectively developing and testing common algorithms that we hoped to one day translate into real domains as scientific problems,” Hasabis said. “We believe that AlphaFold is the first proof of this thesis. These algorithms are now becoming mature enough and powerful enough to be applicable to truly challenging scientific problems.”

DeepMind also entered a CASP protein folding competition in 2018. Although its results at the time were impressive, John Jumper, head of AlphaFold at DeepMind, said the team knew it was some way to produce something with “Really strong biological significance or to be competitive with the experiment. “

This year’s race, however, was no ordinary sailing, and Jumper said DeepMind was gone for three months without making any progress. “Are we going to sit there and worry about running out of data?” He said.

Even as the race approached, Jumper and his team still worried they might have made mistakes. “There can always be a mistake that creeps into machine learning systems,” he said.

But their efforts seem to be working. “We really think we’ve built a system that provides accurate and useful information for experimental biologists,” he said. “The reason you have a structure is to understand something about the natural world and then ask even more questions. We think we’ve built a system that will really help people do that.”


Source link