Without code for DeepMind’s protein AI, this lab wrote its own

Unbeknownst to them, to DeepMind, a detailed scientific article detailing their system was already being reviewed at Nature, according to John Jumper, who heads Project AlphaFold. DeepMind had submitted his manuscript to Nature May 11.

At this point, the scientific community knew little about the DeepMind timeline. That changed three days after Baker’s preprint was published on June 18 when DeepMind CEO Demis Hassabis took to Twitter. “We have worked hard on our full methods document (currently under review) with the accompanying open source code and on providing broad free access to AlphaFold for the scientific community,” he wrote. “More very soon!

On July 15, the same day Baker’s RoseTTAFold article was published, Nature released the unedited but peer reviewed version of DeepMind AlphaFold2 manuscript. Simultaneously, DeepMind created the code for AlphaFold2 available for free on github. And a week later, the team published a huge database of 350,000 protein structures that had been predicted by their method. The revolutionary protein prediction tool, and a vast volume of its predictions, were finally in the hands of the scientific community.

According to Jumper, there is a trivial reason why the DeepMind article and code was not released until more than seven months after the CASP presentation: “We weren’t ready to open the source code or publish. this extremely detailed article that day, ”he says. . Once the document was submitted in May and the team was working on the peer review process, Jumper says they tried to get the document out as soon as possible. “Honestly, we pushed as fast as we could,” he says.

The DeepMind team manuscript was published via NatureExpedited article preview workflow, which the journal uses most frequently for Covid-19 articles. In a statement to WIRED, a spokesperson for Nature wrote that this process is intended as “a service to our authors and readers, with the aim of making particularly noteworthy and urgent peer-reviewed research available as quickly as possible”.

Jumper and Pushmeet Kohli, head of the DeepMind science team, questioned whether Baker’s article took into account the timing of their Nature publication. “From our perspective, we contributed and submitted the document in May, so it wasn’t in our hands, in a sense,” Kohli said.

But CASP organizer Moult believes the work of the University of Washington team may have helped scientists at DeepMind convince their parent company to make their research available for free over a shorter period of time. “I feel like I know them – they are truly exceptional scientists – is that they would like to be as open as possible,” says Moult. “There’s a certain tension there, in that it’s a business venture, and at the end of the day, it has to make money somehow.” The company that owns DeepMind, Alphabet, has the fourth highest market capitalization in the world.

Hassabis calls the release of AlphaFold2 a benefit to both the scientific community and to Alphabet. “This is all open science and we are giving that to humanity, unconditionally, to the system, to the code and to the database,” he said in an interview with WIRED. When asked if there had been any discussion about keeping the code private for business reasons, he replied, “That’s a good question about how we generate value. Value can be provided in different ways, right? One is obviously commercial, but there is also prestige.

Baker does not hesitate to congratulate the DeepMind team for the thoroughness of their publication of paper and code. In a sense, he says, RoseTTAFold was a hedge against the possibility that DeepMind was not acting in a spirit of scientific collaboration. “If they had been less enlightened and had decided not to [release the code], then there would at least have been a starting point that the world could build on, ”he says.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *