Use many agents to sift a biology dataset
Run lots of focused agents over one large biology dataset to surface testable hypotheses faster than one person could.
- What they did
- OpenAI, PerturbAI, and Scripps Research divided analysis of an existing CRISPR map of mouse brain cells among thousands of GPT-Rosalind agents. Each one handled a single gene knockdown, examined how cells reacted, checked prior studies, and drafted a possible explanation. The individual reports were then assembled into a searchable reasoning atlas, and blinded experts evaluated the output.
- How it went
- The full job finished in 32 minutes, while a person was estimated to need months or even a year. Experts picked the distributed reports over a lone agent’s results 75% of the time, and the system highlighted leads involving Tsc1/Tsc2 and DCTN1-linked cholesterol. Those leads still need wet-lab follow-up.
- Tips
- This worked better because the workload was split by perturbation; one agent on the same assignment found only 11 references, compared with more than 700 from the distributed setup. If you try this pattern, keep a human check before you treat any lead as confirmed.
Try it yourself with Claude Code
I have a biological dataset in [folder/path] tied to [research question]. Divide the job across many agents so each one examines one gene, perturbation, sample group, or cell population at a time. For every slice, inspect the data, look for patterns or unusual responses, search the literature, and suggest possible mechanisms, with evidence and caveats. Stop and ask me to approve the plan before any step that would write files, send results, delete anything, submit a job, or make other irreversible changes. Then combine the outputs into a searchable table or atlas and rank the strongest leads with citations, methods, and limitations. Save the scripts and outputs so I can rerun the analysis.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Search for room-temperature magnetic semiconductor candidates
A Claude Opus 5.5 agent team screened Luttinger-compensated magnets and surfaced two predicted room-temperature magnetic semiconductor candidates.
Set up behaviors for a small ROS2 robot
A builder used an onboard agentic OS to program a compact robot through behaviors, code tools, and ROS2 topics.
Use Ghidra tools to map a binary faster
An MCP bridge let an agent decompile code, trace relationships, and rename symbols inside Ghidra from Claude or Cursor.
Use agents to probe open mathematical questions
A group built a math environment where agents chose their own leads, ran tests, and shared findings without a central planner.
Turn a feature phone into a tiny AI controller
They loaded an AI-driven chat tool onto a Nokia 110 and used it to carry out a few simple device actions.
Use an agent to draft parametric geometry systems
A Rhino plugin let a coding agent turn a brief into editable geometry logic stored alongside the project files.