Build a target dossier from gene name to structure to cancer dependency

For a candidate gene, return one page of evidence: disease associations and tractability from Open Targets, protein annotation and PTMs from UniProt, an AlphaFold predicted structure with confidence regions flagged, and CRISPR essentiality plus drug-sensitivity context from DepMap.

   
Problem class Knowledge synthesis
Subject areas Molecular and Cellular Biology, Drug Repurposing and Discovery, Translational Medicine
Evidence level Proposed
Complexity Multi-tool harness
Availability Fully open
Compute Laptop

Problem

Target dossiers are the bread-and-butter of early discovery and translational research. A wet-lab biologist hears a gene name in a seminar, or a computational biologist’s pipeline surfaces a new hit, and the next question is always the same: what do we already know? That means stitching together Open Targets (disease evidence, tractability, mechanism), UniProt (sequence, domains, PTMs, subcellular localization), AlphaFold (predicted structure, low-confidence loops to avoid in modelling), and DepMap (CRISPR essentiality across cancer cell lines, optional drug-sensitivity context). Each lookup is fast on its own; the cost is context-switching across four portals and copy-pasting evidence into a doc. Solved looks like: one prompt, one page of cited evidence, in under five minutes.

  1. Install the four components.

    /plugin marketplace add anthropics/life-sciences
/plugin install open-targets@life-sciences
/plugin marketplace add K-Dense-AI/claude-scientific-skills
/plugin install gget@claude-scientific-skills
/plugin install depmap@claude-scientific-skills

    Add the UniProt and AlphaFold MCP servers per their catalog pages — both are stdio Node servers, no auth needed.

  2. Drive the dossier with a single multi-step prompt. A minimal version:

    Build a one-page target dossier for STK11. Use:
- open-targets: top 5 disease associations by association score,
  plus tractability evidence (clinical compounds, antibodies, small
  molecules).
- uniprot: canonical isoform, domain architecture, PTMs,
  subcellular localization, key paralogs.
- alphafold: predicted full-length structure; flag any region with
  pLDDT < 70 as low-confidence.
- depmap: CRISPR essentiality (Chronos) summary across all
  lineages and any lineage where the gene is selectively essential.
Cite each fact with the source database and an accession or query.
Render as a single Markdown page under dossiers/STK11.md.

  3. Sanity-check the cross-references. UniProt’s accession should match the gene symbol Open Targets used; AlphaFold’s structure should be keyed on that same UniProt accession; DepMap should use the HUGO gene symbol or Entrez ID. Mismatches usually mean the gene has paralogs (e.g., the GAPDH / GAPDHS family) — re-prompt with the explicit accession.

  4. Iterate. Once the first dossier is right, save the prompt as a slash command (e.g., /dossier) and parameterize on the gene symbol. The same prompt will work for any future gene without re-explaining the structure.

Why this assembly

Rung 3 of the simplicity ladder. The problem genuinely requires four heterogeneous data sources — disease evidence, protein annotation, structure, and dependency. No single MCP covers all four. A toolbelt of four read-only components is still well within the rung-3 budget (≤ 3 tightly-coupled components is the soft target; here Open Targets + UniProt + AlphaFold + DepMap is four loosely-coupled lookups orchestrated by a single prompt, which the agent loop handles cleanly). Escalating to an autonomous-science system (Biomni) would work but adds loop overhead with no extra capability — the dossier is a knowledge-synthesis problem, not a hypothesis-generation problem.

Availability

Fully open. Open Targets data is CC0; UniProt, AlphaFold, and DepMap are free for academic use. All four MCP/skill components are OSS. No subscription or institutional account is required for the public data shown here. (DepMap downloads can be hundreds of MB on first run; cache locally.)

Compute requirements

Laptop-sufficient. All four components are read-only API or file lookups; the only heavy step is the initial DepMap data download (~200 MB for the standard files), cached after first use. No GPU needed.

Evidence

Proposed. No published end-to-end benchmark of this exact four-component assembly is known. The closest documented analogue is the Biomni paper’s case studies, where the same agent integrates dozens of biomedical databases (including Open Targets, UniProt-class annotation, and AlphaFold) to answer multi-step biomedical questions and matches expert performance on LAB-Bench DbQA (74.4% vs 74.7% human experts, Huang et al., 2025). Each component in this recipe has independent evidence (Open Targets is the consortium-maintained reference for target-disease associations; UniProt is the canonical protein annotation; AlphaFold confidence-based filtering is standard practice; DepMap essentiality scoring follows Behan et al. 2019 and Dempster et al. 2021). The composition has not been benchmarked.

Alternatives considered

  • gget alone. The gget skill covers Ensembl, UniProt, and PDB lookups in one interface. Reach for gget when the dossier is gene-and-protein-centric only and you don’t need DepMap dependencies or Open Targets disease evidence. Cheaper but less complete.
  • Open Targets alone. Open Targets exposes a lot through GraphQL, including some of the cross-reference data UniProt provides. Reach for Open-Targets-only when target-disease evidence is the primary question and structure/dependency are nice-to-have.
  • Biomni. An autonomous-science system that already has Open Targets, UniProt-class annotation, AlphaFold, and DepMap-style data wired in. Reach for Biomni when the dossier is one step of a larger autonomous pipeline (e.g., generate target list → build dossiers → propose experiments). For a one-shot dossier, the rung-3 toolbelt is simpler and more transparent.

See also

Sources

Tried this recipe?

Share feedback — what worked, what didn’t, what you’d change. The form opens with this recipe pre-selected and a link back to this page.