Add ReactomeGSA tools: Reactome does do gene set analysis

[adamjohnwright]

Reactome has two analysis services. This server knew about one.

AnalysisService	over-representation over a list of identifiers	already here
ReactomeGSA (gsa.reactome.org)	gene set analysis over an expression matrix — PADOG, Camera, ssGSEA, terapadog	new

Camera is described by the service itself as "a gene set analysis algorithm similar to the classical GSEA algorithm".

How the gap was found

A Reactome chatbot, asked "I want to run a gsea with my list of genes", replied that the information was not available in the Reactome Knowledgebase — then offered fgsea, bigomics.ch and a YouTube tutorial.

Reactome can do this. Nothing here could reach the service that does it, so a Reactome user was sent to someone else's tools.

Pointer to its existence: the Galaxy CLI (reactome/reactome_galaxy), which wraps both services.

Five read-only tools

reactome_gsa_methods, reactome_gsa_data_types, reactome_gsa_search_datasets, reactome_gsa_examples, reactome_gsa_sources.

The descriptions carry the distinction that caused the confusion, because a description is the whole of what a model reads before choosing. Gene set analysis needs an expression matrix with sample groups; a user holding a list of gene names wants over-representation, whatever they called it. Two of the tools say so explicitly and name reactome_analyze_identifiers as the alternative.

Submitting an analysis is deliberately absent

POST /analysis takes the entire expression matrix inline — not something a chat user can paste, and not something to push through a tool result against a 40,000-character cap.

What is reachable is covered instead. ReactomeGSA loads from Expression Atlas, Single Cell Expression Atlas, GREIN and GEO, so reactome_gsa_search_datasets lets someone with no data of their own be pointed at a published dataset:

## Public datasets matching "melanoma" in Homo sapiens
**Found:** 100

### RNA-seq melanoma
**ID:** GSE50535 · **Species:** Homo sapiens · **Source:** GREIN

Actually running one is the web interface, the Galaxy tool, or the R package — and every tool says so.

A trap worth the comment it carries

GSA's species filter wants the name:

species=9606           -> 0 results     (silently)
species=Homo sapiens   -> 100 results

The exact opposite of the Content Service's eventsHierarchy, which wants the id and answers HTTP 500 for the name. Both the tool description and the empty-result message say which, because a filter that silently returns nothing is the failure mode this repo has spent the week removing.

Verification

• 53 → 58 tools, 90 → 101 tests, npm run check green
• sweep covers all 58 with 21 content expectations, exit 0, no suspicious output
• every fixture copied from the live service, with the endpoint recorded above the type
• coverage thresholds raised to match (57% statements)

🤖 Generated with Claude Code