How scientists are using Claude to accelerate research and discovery

Source: Anthropic Date: 2026-01-15 URL: https://www.anthropic.com/news/accelerating-scientific-research

Summary

Anthropic published research use cases: Biomni (Stanford) — Claude-powered agent across 25+ biological subfields, genome-wide association study in 20 minutes vs. months; MozzareLLM (MIT) — automated gene knockout experiment interpretation; Lundberg Lab (Stanford) — Claude generating gene candidates from molecular properties rather than literature guessing. Claude for Life Sciences connectors launched October 2025. Claude Opus 4.5 improvements in figure interpretation, computational biology, protein understanding. Partnerships: AI for Science program, Undiagnosed Diseases Network.

Implications

• Science / proof of concept from Stanford + MIT. Stanford’s Biomni and MIT’s MozzareLLM are credible academic implementations — not vendor-created demos but researcher-built tools using Claude. The 20-minute GWAS (vs. months) is the most dramatic compression ratio claimed in any Anthropic science use case.
• Genome-wide association study compression. A GWAS identifies genetic variants linked to diseases — it’s computationally intensive and requires weeks/months of analysis, human judgment, and literature review. Compressing this to 20 minutes via Claude agent doesn’t mean the science is done faster; it means the computational and analysis phases that previously required human oversight are now automated.
• Gene candidate generation from molecular properties. The Lundberg Lab use case (Claude suggesting gene candidates based on molecular properties rather than literature-based guessing) represents AI-driven hypothesis generation — the most ambitious scientific automation use case Anthropic had published.
• Undiagnosed Diseases Network. The rare disease diagnosis partnership is both commercially marginal and scientifically significant — rare diseases (8,000+ known) lack clinical expertise, making AI assistance genuinely needed. It’s also the use case with the most compelling human interest narrative.
• Watch: whether Biomni’s 20-minute GWAS results were independently validated; how the AI for Science program credit usage tracked against actual publications; whether Claude Opus 4.5’s protein understanding advances translated into drug discovery applications.