Gemini for Science day 2: AlphaFold + AlphaGenome pitched to 'solve all diseases'

Day two of Google I/O coverage shifted from the Gemini Omni / Spark consumer launches to Gemini for Science, the research-oriented stack The Verge profiled in a column on May 20. The pitch: AlphaFold and AlphaGenome — DeepMind's structural biology and genomics models — are now first-class tools inside the Gemini ecosystem, exposed both to internal Isomorphic Labs drug-discovery work and to outside researchers via Google's experimental 'Hypothesis Generation' tool and the DeepMind Co-Scientist multi-agent system announced this week.
The technical hook is integration depth. AlphaFold's protein structures and AlphaGenome's variant-effect predictions are being piped into Gemini as callable tools, letting researchers ask natural-language questions ('which variants in this gene are most likely pathogenic given this structure?') and get grounded answers backed by the underlying models. Co-Scientist layers a multi-agent loop on top — a hypothesis generator, a critic, and an experimentalist agent that proposes wet-lab validations.
The skeptical take, which The Verge foregrounded, is the gap between structural-biology demonstrations and clinical impact. AlphaFold has unquestionably accelerated structural biology, but the drug-discovery pipeline from target to approval still runs 7-12 years and remains gated by trials, manufacturing, and regulatory science — none of which Gemini for Science touches. Pichai's 'solve all diseases' framing landed as marketing, not science, even among biotech-friendly observers.
Competitively this is Google's strongest story this week. OpenAI's discrete-geometry result is a single math claim; Anthropic just hired Karpathy. Google has actual deployed scientific tooling with track-record models (AlphaFold won a Nobel) plus the infrastructure to run inference at scale. The risk is that the consumer story (Omni, Spark) and the science story compete for narrative oxygen, leaving both undersold. Watch for which Gemini-for-Science partner labs publish first results.