DeepMind Has Reassigned the Original AlphaFold Team Across Gemini, Biology, and Isomorphic Labs
DeepMind Has Reassigned the Original AlphaFold Team Across Gemini, Biology, and Isomorphic Labs
On July 29, the Financial Times reported that, over the past year, Google DeepMind has reassigned most of the original authors of the AlphaFold papers to projects involving Gemini, enzyme design, genomics, nuclear fusion, and Isomorphic Labs. About a quarter of the employed authors of the early AlphaFold papers have left the company.
AlphaFold took a protein's amino acid sequence and predicted its three-dimensional structure. A researcher could use this prediction as a starting point for selecting a drug target or planning an experiment. AlphaFold had one straightforward measure of success: how closely its prediction matched the protein's actual structure.
The team worked on this problem for years, after which DeepMind made a database of predicted structures available to other laboratories. John Jumper and Demis Hassabis received half of the 2024 Nobel Prize in Chemistry for protein structure prediction. AlphaFold showed how a single specialized team could turn a difficult problem in biology into a tool that researchers could use.
DeepMind is now distributing this expertise. Some employees have moved to Isomorphic Labs, an Alphabet company that develops models for drug discovery. Others are working on enzyme design, genomics, and nuclear fusion. Instead of maintaining one team focused on one problem, DeepMind is building several research programs in which AlphaFold methods can work alongside Gemini.
Pushmeet Kohli, head of AI for Science at DeepMind, told the FT that Gemini-based systems should assist scientists and automate specific parts of the research process. Google Research has already described an AI co-scientist built on Gemini 2.0: a researcher defines a goal, the system proposes hypotheses, a literature review, and a research plan, and then incorporates new ideas and feedback from the scientist.
This new organization has a different measure of success: specific hypotheses, molecules, and experiments that laboratories can test. Those results will show what DeepMind gained by reorganizing the AlphaFold team.