Without DeepMind’s Protein AI code, this lab has written its own


Unaware of them, DeepMind has already reviewed extensive scientific work detailing their system. Nature, according to John Jumper, who runs the AlphaFold project. DeepMind submitted his manuscript Nature May 11.

At the time, the scientific community knew little about DeepMind’s timeline. That changed three days after Baker’s preprint became available, on June 18, when DeepMind CEO Demis Hassabis went on Twitter. “We have been working slowly to develop our full method paper (currently under review) with accompanying open source and to provide broad free access to AlphaFold to the scientific community,” he wrote. “More soon!”

On July 15, the same day Baker’s work RoseTTAFold was published, Nature released by DeepMind unedited but peer-reviewed AlphaFold2 Manuscript. At the same time, DeepMind created code for AlphaFold2 freely available on githubu. And a week later, the team released an huge database of the 350,000 protein structures predicted by their method. The revolutionary protein prediction tool and a large number of its predictions were finally in the hands of the scientific community.

According to Jumper, there is a trivial reason why DeepMind’s paper and code were not published until more than seven months after CASP’s presentation: “That day we were not ready to open the source or to publish this extremely detailed article,” he says. After the paper was submitted in May and the team was working on the review process, Jumper says they tried to pull out the paper as soon as possible. “Honestly, we pushed as fast as we could,” he says.

The DeepMind team manuscript was published via Nature‘Accelerated Workflow for Reviewing Articles’, which the journal most commonly uses for papers on Covid-19. In a statement to WIRED, a spokesman Nature wrote that this process was conceived as “a service to our authors and readers, in the interest of making particularly valuable and time-sensitive peer-reviewed research available as soon as possible”.

Jumper and Pushmeet Kohli, leaders of DeepMind’s science team, discussed whether Baker’s paper was counted at the time of their Nature publication. “From our perspective, we gave our contribution and handed over the work in May, so in a sense it was out of our hands,” says Kohli.

But CASP organizer Moult believes the work of the University of Washington team may have helped DeepMind scientists convince their parent company to make their research available in a shorter period of time. “I feel like I know them – they’re really outstanding scientists – that they want to be as open as possible,” says Moult. “There’s a certain tension there, because it’s a commercial enterprise, and in the end you have to make some money somehow.” The company that owns DeepMind, Alphabet, has the fourth largest market capitalization in the world.

Hassabis characterizes the release of AlphaFold2 as a benefit for both the scientific community and the Alphabet. “This is all open science and we are giving it to humanity, without any connections, systems, codes and databases,” he said in an interview with WIRED. Asked if there is a debate about preserving code privacy for commercial reasons, he said: “It is a good question how we deliver value. Value can be delivered in many different ways, right? One is obviously commercial, but there is also prestige. ”

Baker quickly praised the DeepMind team for the thoroughness of their work and publishing the code. In a sense, he says, RoseTTAFold was a protection against the possibility that DeepMind would not operate in a spirit of scientific collaboration. “That they were less enlightened and decided not to do it [release the code], then the world would at least be a starting point for upgrading, ”he says.

Source link


Please enter your comment!
Please enter your name here