There is no DeepMind protein AI code, this laboratory wrote it by itself


They didn’t know that at DeepMind, an extensive scientific paper detailing their system was already under review. natureAccording to John Jumper, who leads the AlphaFold project. DeepMind has submitted their manuscript to nature May 11th.

At that time, the scientific community knew very little about DeepMind’s timeline. Three days after the release of Baker’s preprint, on June 18, DeepMind CEO Demis Hassabis visited Twitter and the situation changed. He wrote: “We have been working hard to complete our complete method paper (currently under review) with open source code and provide the scientific community with extensive free access to AlphaFold.” “There will be more soon!”

On July 15, the same day Baker’s RoseTTAFold paper was published, nature Released DeepMind’s unedited but peer-reviewed AlphaFold2 manuscriptAt the same time, DeepMind made code for AlphaFold2 Free of charge On github.A week later, the team Released One Huge database Their method predicted 350,000 protein structures. The revolutionary protein prediction tool and its large number of predictions are finally in the hands of the scientific community.

According to Jumper, the reason why DeepMind’s papers and code were not released until more than seven months after the CASP presentation was common: “We were not ready to open source or publish this extremely detailed paper that day,” he said. Submitted in May After the paper, the team is completing the peer review process, and Jumper said they are trying to publish the paper as soon as possible. “To be honest, we have been pushing as fast as possible,” he said.

The manuscript of the DeepMind team was published in natureThe accelerated article preview workflow of the journal is most commonly used for Covid-19 papers. In a statement to Wired magazine, nature Writes that this process aims to “provide services for our authors and readers, with the aim of providing particularly noteworthy and time-sensitive peer-reviewed research as soon as possible.”

Pushmeet Kohli, head of the Jumper and DeepMind scientific team, disputed whether Baker’s paper affected the timing of their research. nature Publication. “From our perspective, we contributed and submitted this paper in May, so in a sense, it is beyond our grasp,” Kohli said.

But CASP organizer Moult believes that the work of the University of Washington team may help DeepMind scientists persuade their parent company to provide their research for free in a shorter period of time. “I understand them-they are very outstanding scientists-and the feeling is that they want to be as open as possible,” said Mort. “There is some tension there, because it is a commercial enterprise, and ultimately it must make money in some way.” Alphabet, the company that owns DeepMind, has the fourth largest market capitalization in the world.

Hassabis described the release of AlphaFold2 as beneficial to both the scientific community and Alphabet. In an interview with WIRED, he said: “This is all open science, and we provide it to humans without any additional conditions, systems, codes, and databases.” When asked if there was a discussion about keeping the code secret for business reasons , He said: “This is a very good question, how do we deliver value. Value can be delivered in many different ways, right? One is obviously commercial, but it also has prestige.”

Baker quickly praised the thoroughness of the DeepMind team’s paper and code release. He said that in a sense, RoseTTAFold is a hedge against the possibility that DeepMind does not act in the spirit of scientific cooperation. “If they are not so enlightened and decide not to [release the code], Then there is at least a starting point for the world to continue to develop,” he said.


Source link