About this template
Target briefs with accessions you can click
If your biology agent invents proteins and never cites a database, this is the fix. This OpenScience box writes target briefs grounded in UniProt, PDB and ChEMBL — with the identifiers printed, so you can open them and check.
What an AI drug discovery brief looks like here
Ask for a target. Get a sequence-to-structure-to-chemistry brief back.
The workflow runs in that order on purpose. UniProt for the canonical sequence and function. PDB for representative structures. ChEMBL and PubChem for known ligand and bioactivity context. Every claim arrives with its accession attached, because the agent's job is to find and frame the evidence — not to be believed.
The first five minutes
- Fork the template, then open Open OpenScience on the VM.
- Set one model key in the shell:
export ANTHROPIC_API_KEY=...
- Ask the biology agent:
Brief EGFR kinase: UniProt entry, representative PDB structures, and ChEMBL ligands — with accessions I can open.
- Open three of the identifiers it hands back.
If the IDs resolve and match what the brief said, the box works. If they don't, you learned that in five minutes instead of five hours.
What lands in your fork
- A comp-bio oriented biology agent for OpenScience
- bio-db-query — UniProt, PDB, ChEMBL and PubChem lookups
- bio-lit-target — literature and target framing
- The OpenScience browser workspace, reachable through Open OpenScience
What you bring (BYOK)
One model key. ANTHROPIC_API_KEY is the path shown above; OPENAI_API_KEY, GEMINI_API_KEY and OPENROUTER_API_KEY are read too.
Keys stay on your microVM, and the template ships scrubbed — no credentials came across from the box it was captured on. Model spend is yours.
Who this is for
- Structural and computational biology work you already do by hand
- You want database-grounded briefs, and you already speak UniProt/PDB/ChEMBL
- You would rather check an accession than trust a paragraph
Honest limits
It queries public databases and frames literature. That is the whole job.
No docking. No ADMET prediction. No molecular dynamics. No wet-lab LIMS or ELN integration on day one. No ML training loops either — those are the ML co-scientist and inference-lab templates, not this one. And a language model can still misread a record, which is exactly why every brief hands you the accession instead of asking for trust.
Fork it. Ask for EGFR. Click the accessions. Your first jurniti purchase carries a 30-day money-back guarantee.