The scientific community is applauding the work of DeepMind, a Google subsidiary, which, thanks to an artificial intelligence model, is now able to predict the structure of proteins with unprecedented accuracy. This will speed up countless research projects.

Google subsidiary, artificial intelligence research lab DeepMind – known for creating AIs that can beat the greatest champions in go or Quake III – has just created a map of human proteins, with all the data freely available. A project made possible by artificial intelligence, the impact of which could be as significant as that of mapping all human genes, according to leading representatives of the scientific community. Venki Ramakrishnan, winner of the 2009 Nobel Prize in Chemistry, said in a statement: “This is an incredible breakthrough, and it has happened much sooner than many experts would have predicted. It will be exciting to see the many ways in which biological research will change radically.

Proteins are long, complex molecules that perform countless functions in the human body, characterized by their shape, structure and the way they fold. Studying proteins and understanding how they fold for their purpose is very useful for scientists, whether it is to scrutinize the functioning of the human body or the development of diseases, not to mention the creation of treatments or drugs, such as against Covid.

Recommended article:

Years of research saved

To save scientists long and costly experiments to determine the shape of proteins (the structure of a single one can currently take months or years), DeepMind has come up with an artificial intelligence model called AlphaFold2, which is able to predict the structure of these large, complex molecules with great accuracy. An AI that learned on its own and grew to its current level of accuracy, allowing DeepMind to publish hundreds of thousands of predictions.

See also  The Logitech G203 LIGHTSYNC gamer mouse on sale

Interviewed by The Verge, Demis Hassabis, co-founder and director of DeepMind, explains that he sees this leap forward as “the culmination of more than 10 years of diverse work.” “All along, this has been the goal we’ve been working towards: to achieve major breakthroughs with AI, test them on games and then apply them to real-world problems to see if we can accelerate scientific research and use them for the benefit of humanity,” he adds.

Last week, the London-based company published in <a href=”https://www.nature.com/articles/d41586-021-02025-4″ target=”_blank” rel=”nofollow noopener”>scientific journal Nature its methodology and the main principles of its algorithms, demonstrating to be able to predict the structure of proteins almost perfectly. In a second paper, DeepMind proved to be able to predict the structure of 60% of the amino acids in the human body, while demonstrating the effectiveness of its AI on other organisms (conclusive tests were conducted on mice, flies and the intestinal bacteria E.coli). This was followed by the posting of structure predictions for 350,000 proteins in a database hosted by theEuropean Bioinformatics Institute‘s molecular biology laboratory.

Edith Heard, director of this laboratory, does not hide her enthusiasm: “The precise prediction of their structures has a wide range of scientific applications, from the development of new drugs and treatments for diseases to the design of future crops capable of resisting climate change, or enzymes capable of degrading plastics […] The applications are limited only by our imagination.

200 million proteins in the sights

From a purely medical point of view, structurally altered proteins can be responsible for the development of diseases such as Alzheimer’s and Parkinson’s. Being able to predict the shape of a protein could help scientists control and alter it to improve its function by changing its DNA sequence, or design drugs that target it directly.

See also  NBA 2K22: Where to pre-order the title at the best price?

At present, some 180,000 proteins have been analysed out of the more than 200 million known in living organisms. For this reason, reliable prediction of their structure is considered one of the greatest challenges in biology. “Our ambition is to extend the database to the entire universe of proteins, more than 200 million proteins, in the coming months,” says Demis Hassabis of DeepMind. The only limitation is that proteins are dynamic molecules that constantly change their structure depending on their environment, whereas DeepMind’s algorithm can only predict a static structure.

This was one of those moments where my hairs stood on end […] We are able to use this information to accelerate the development of enzymes that can digest plastic. We’ve just gained several years of work,” says John McGeehan, director of an enzyme research unit at the University of Portsmouth, who has been testing AlphaFold in recent months. He also praises the availability of DeepMind’s work in open source, accessible to the greatest number of people, which will certainly accelerate scientific research in many fields of application. And earn Google a Nobel Prize through DeepMind?

Sources: DeepMind, <a href=”https://www.ft.com/content/fbcc9af4-8dcd-4385-85d5-59c180175b67″ target=”_blank” rel=”nofollow noopener”>The Financial Times, The Verge