The team responsible for one of artificial intelligence's cleanest scientific wins has stopped existing as a team. Google DeepMind confirmed to the Financial Times on July 29 that it dissolved the dedicated group behind AlphaFold, the protein structure prediction system that earned Demis Hassabis and John Jumper the 2024 Nobel Prize in Chemistry. Most of the researchers who authored the original AlphaFold papers spent the past year moving to other assignments inside the company. Close to a quarter of them left Google DeepMind altogether.

The team that won a Nobel Prize doesn't exist anymore

AlphaFold started in 2018 as a bet on a single unsolved problem: predict a protein's three-dimensional shape from its amino acid sequence, a task that had taken structural biologists years of crystallography and electron microscopy per structure. It worked. It won a Nobel Prize. And the organizational model that produced it, a dedicated team pointed at one hard problem until it broke, is what Google DeepMind just walked away from.

Pushmeet Kohli, the lab's vice president of research, told the Financial Times the strategy has evolved. Nine years of grand-challenge teams, each with a concrete goal, gave way to something else: Gemini as the coordinating layer across scientific work rather than a separate unit for each discovery.

The reassignments reveal more than the departures

Reassignment, not departure, accounts for most of the AlphaFold team's disappearance. Researchers moved into Gemini-focused projects, enzyme design, nuclear fusion modeling, and genomics work. Others transferred to Isomorphic Labs, the Alphabet-owned drug discovery company spun out of DeepMind in 2021 that already holds partnerships with Novartis and Eli Lilly.

That is the tell.

A company retiring a research line moves people to unrelated work. A company folding a research line into a broader platform moves people onto the platform. Google chose the second path, and it chose it for the exact team that gave the lab its Nobel Prize.

"The strategy has evolved."
Pushmeet Kohli, VP of Research, Google DeepMind

The database survives. The team that built it does not.

Nothing about the AlphaFold Server, AlphaFold 3, or the AlphaFold Database itself has been discontinued. The database still holds more than 200 million protein structure predictions, free to academic and commercial researchers. By DeepMind's own count as of last November, over 3 million researchers across more than 190 countries had used it, with roughly a third of that research aimed at understanding disease.

The commercial edge of that work has already moved past AlphaFold's public numbers. Isomorphic Labs said in February that its own drug discovery engine, IsoDDE, had more than doubled AlphaFold 3's accuracy on a generalization benchmark for protein-ligand structures, the kind of prediction that matters most for finding a drug that actually binds its target. I covered Isomorphic's progress on that front in April. The asset base AlphaFold built keeps compounding. The team credited with building it does not exist to keep pace with its own commercial spinoff.

Anthropic's biology bet just picked up three co-authors

Jumper announced in June he was leaving DeepMind for Anthropic. Two of his AlphaFold colleagues followed him. One DeepMind employee told the Financial Times the three researchers were considered central to the team, and their exit caused surprise internally.

The timing lines up with Anthropic's own life-sciences push. The company launched Claude Science on June 30, a workbench built on its existing Claude models rather than a dedicated biology model, connecting researchers to more than 60 skills and databases spanning genomics, proteomics, structural biology, and cheminformatics. Anthropic has also started an internal drug discovery program aimed at diseases traditional pharmaceutical companies skip for lack of commercial upside. Hiring the researchers behind the field's most cited breakthrough, three months into that push, is not a coincidence a hiring manager would need explained.

Unknowns and uncertainties

Several questions here do not have public answers yet. Whether the AlphaFold Database gets the same update cadence without a dedicated team, or whether Isomorphic Labs becomes its de facto successor for anything beyond the existing public release, is not something DeepMind has addressed. Whether Google's next scientific breakthrough comes from a Gemini-coordinated effort or from a team structure it just dismantled is an open bet, not a settled one. And whether the three researchers now at Anthropic are building toward a public research release or staying inside Claude Science's commercial workbench is not yet clear from either company's public statements.

About Shashi Bellamkonda

Loading bio…

LinkedIn Profile

Get new posts by email:

Subscribe

Powered by