AlphaFold’s Standalone Team Is Gone as Google Reorganizes Around Gemini

Lead

The group behind AlphaFold, one of the most influential AI-for-science projects to date, is no longer functioning as an independent team inside Google DeepMind. According to the report, the researchers who once focused tightly on protein structure prediction have been redistributed across Google and Alphabet projects, while some prominent members have left the company altogether.

Key points

• A structural reorganization, not a simple cancellation: The AlphaFold effort has not disappeared, but the original standalone team has been dissolved. Former members are now working on Gemini, AI coding, genomics, protein and enzyme design, fusion research and other scientific programs.
• Google frames it as expansion: Pushmeet Kohli, Google Cloud chief scientist and DeepMind vice president of science, said DeepMind’s scientific work has not shifted away from its mission. Instead, he described the scope as expanding into protein function, genomics and Gemini-powered scientific agents.
• High-profile departures matter: John Jumper, who led AlphaFold 2 and later shared the 2024 Nobel Prize in Chemistry with Demis Hassabis for work on protein structure prediction, has left DeepMind for Anthropic. Jonas Adler and Alexander Pritzel, both connected to AlphaFold research, are also reported to be joining Anthropic.
• Alphabet is pushing toward drug discovery: Some former AlphaFold researchers have moved to Isomorphic Labs, the Alphabet company created to apply AlphaFold-like capabilities to drug discovery and design.

Why it matters

AlphaFold succeeded through a very specific research model: a small interdisciplinary team spent years attacking one hard, narrow scientific problem. Machine learning experts, biologists and engineers concentrated on whether protein structures could be predicted from amino acid sequences with high accuracy. That model produced not just papers, but a tool that reshaped structural biology.

Google’s current strategy appears different. Rather than building a dedicated team for each scientific challenge, it is trying to make Gemini a shared foundation for scientific work. Tools such as AI Co-Scientist, AlphaEvolve and Science Skills point toward a platform approach in which models, data resources and agent workflows can be reused across biology, materials, mathematics and fusion research.

The open question is whether such a platform can reproduce the depth of focus that made AlphaFold possible. Spreading experts across more projects may increase their reach, but it may also reduce sustained attention on any single grand challenge. The AlphaFold team’s dissolution is therefore less a story about one project ending than about Google’s changing theory of how major scientific AI breakthroughs should be organized.

Source: 量子位