research-article-analysis

name

research-article-analysis

description

Systematically validate a research article's claims against TogoMCP RDF databases — molecular formulas, reaction equations, pathways, protein function, GO definitions — instead of trusting the paper's text or keyword-search snippets. Use whenever the user asks to "analyze this paper/article", fact-check or validate a study's biology, verify claimed metabolites/reactions/pathways/ proteins/processes against databases, score the evidence for a paper's claims, or build a cross-database evidence chain (ChEBI → Rhea → UniProt → Reactome → GO). Triggers on "article analysis", "validate this study", "check this paper's claims". The full method lives in references/research_article_analysis.md.

Research Article Analysis (claim validation via TogoMCP SPARQL)

Validate a paper claim-by-claim by running structured SPARQL against five RDF databases — ChEBI, Rhea, Reactome, UniProt, GO — and citing the query results (exact formulas, equations, EC numbers, GO definitions), not the paper's prose or keyword-search text. The driver is the TogoMCP MCP tools (run_sparql, the search_* tools, get_MIE_file).

The full phased checklist, ID-tracking templates, scoring rubric, and anti-skipping safeguards live in references/research_article_analysis.md — read it for the deliverable structure and discipline. This SKILL.md is the operational path: the SPARQL that actually works (the printed templates have two bugs — see Gotchas), the verified tool calls, and the traps.

Prerequisites

• The TogoMCP MCP server connected to your client (e.g. Claude Desktop, Claude Code). Its run_sparql, search_*, and get_MIE_file tools are what drive this skill.
• Network access from that server to rdfportal.org.

Warm-up check (confirm connectivity before trusting the templates)

Before relying on the templates, send the ChEBI structure query below through the run_sparql tool and confirm it returns a row — that proves the endpoint is reachable and the (chemrof-namespace) pattern still resolves. If it returns nothing, that database's query type is broken (endpoint down or schema drift) — re-read its MIE via get_MIE_file and fix the pattern below before relying on it.

Run the analysis (phases, with verified queries)

Follow the phase discipline in the reference doc: 2A select databases (5 rules), 2B get_MIE_file(database=...) for each, 2C keyword-search to get IDs, 2D run 2+ SPARQL per database using those IDs, 3A reconstruct the DB-backed logical flow (results → conclusion) + 3B triage the gaps, 4 score flow soundness, then deliver the four-part FINAL SUMMARY (main claim / conclusion / DB-backed flow / gap analysis). Drive 2D with the TogoMCP run_sparql tool. The query types below were all run live (bile-acid example) and returned data — copy these, not the reference doc's printed templates.

ChEBI — structure (⚠ uses chemrof:, NOT chebi:):

run_sparql(database="chebi", sparql_query="""
PREFIX obo: <http://purl.obolibrary.org/obo/>
PREFIX chemrof: <https://w3id.org/chemrof/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT ?label ?formula ?mass ?smiles ?inchikey
FROM <http://rdf.ebi.ac.uk/dataset/chebi> WHERE {
  obo:CHEBI_16359 rdfs:label ?label ;
    chemrof:generalized_empirical_formula ?formula ; chemrof:mass ?mass .
  OPTIONAL { obo:CHEBI_16359 chemrof:smiles_string ?smiles }
  OPTIONAL { obo:CHEBI_16359 chemrof:inchi_key_string ?inchikey }
}""")
→ cholic acid, C24H40O5, 408.579, SMILES, InChIKey BHQCQFFYRZLCQQ-OELDTZBJSA-N

ChEBI hierarchy uses rdfs:subClassOf + the obo/CHEBI_ filter (works as printed in the reference doc). Definition is obo:IAO_0000115; monoisotopic mass is chemrof:monoisotopic_mass; InChI string is chemrof:inchi_string.

Rhea — equation search & participants (work as printed):

run_sparql(database="rhea", sparql_query="""
PREFIX rhea: <http://rdf.rhea-db.org/>
PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>
SELECT ?reaction ?equation ?status ?ec WHERE {
  ?reaction rdfs:subClassOf rhea:Reaction ; rhea:equation ?equation ; rhea:status ?status .
  FILTER(CONTAINS(LCASE(?equation), "cholate"))
  OPTIONAL { ?reaction rhea:ec ?ec }
} LIMIT 10""")
→ RHEA:14541 "choloyl-CoA + H2O = cholate + CoA + H(+)", EC 3.1.2.27

Participants: VALUES ?reaction { <http://rdf.rhea-db.org/14541> } then rhea:side → rhea:contains → rhea:compound → rhea:chebi → links to ChEBI IDs.

Reactome — pathway structure & complex (work as printed): always FROM <http://rdf.ebi.ac.uk/dataset/reactome>; FILTER(CONTAINS(LCASE(?displayName), "bile acid")) for pathways, ?complex a bp:Complex + bp:component for complexes (e.g. "GPBAR1: Bile acids"). For UniProt xrefs inside Reactome use bp:db "UniProt"^^xsd:string.

UniProt — function & GO (⚠ split into TWO queries):

# Query A: name + EC. Do NOT add up:classifiedWith here — the join TIMES OUT.
run_sparql(database="uniprot", sparql_query="""
PREFIX up: <http://purl.uniprot.org/core/>
PREFIX uniprot: <http://purl.uniprot.org/uniprot/>
SELECT ?mnemonic ?fullName ?ecNumber WHERE {
  uniprot:Q8TDU6 up:reviewed 1 ; up:mnemonic ?mnemonic ; up:recommendedName ?name .
  ?name up:fullName ?fullName .
  OPTIONAL { uniprot:Q8TDU6 up:enzyme ?ecNumber }
} LIMIT 10""")
→ GPBAR_HUMAN, "G-protein coupled bile acid receptor 1"

# Query B: GO terms in a separate query (fast on its own).
run_sparql(database="uniprot", sparql_query="""
PREFIX up: <http://purl.uniprot.org/core/>
PREFIX uniprot: <http://purl.uniprot.org/uniprot/>
SELECT DISTINCT ?goTerm WHERE {
  uniprot:Q8TDU6 up:classifiedWith ?goTerm .
  FILTER(STRSTARTS(STR(?goTerm), "http://purl.obolibrary.org/obo/GO_"))
} LIMIT 50""")
→ GO_0005886, GO_0038182, GO_0097009, ...

GO — definitions & hierarchy (work as printed): always FROM <http://rdfportal.org/ontology/go>, DISTINCT, and the obo/GO_ filter. Definitions via obo:IAO_0000115; hierarchy via rdfs:subClassOf.

Cross-database evidence chain (the reference doc's required deliverable) is real and reproducible: ChEBI cholate (CHEBI_15378) appears as a rhea:chebi participant of RHEA:14541, whose EC links to a UniProt enzyme, which appears in the Reactome "GPBAR1: Bile acids" complex, annotated with GO terms.

Read MIE files first for any database you haven't queried this session: get_MIE_file(database="chebi"), etc. The MIE is the authoritative, freshest schema source.

Gotchas (the template bugs and silent-fail traps)

• ChEBI: the reference doc's printed structure query returns ZERO rows. It uses chebi:formula / chebi:mass / chebi:smiles / chebi:inchikey (http://purl.obolibrary.org/obo/chebi/). RDF Portal's ChEBI stores those under the chemrof namespace instead: chemrof:generalized_empirical_formula, chemrof:mass, chemrof:smiles_string, chemrof:inchi_key_string (https://w3id.org/chemrof/). Use the corrected query above. The hierarchy query is fine because it only uses rdfs:subClassOf.
• UniProt: combining up:recommendedName + up:classifiedWith in one query TIMES OUT (60s) on the SIB endpoint even with a single bound accession. Split: name/EC in one query, GO terms (classifiedWith) in another. The workflow's "Query Type 1" template combines them — don't.
• run_sparql database= vs endpoint_name= are different axes (single DB key vs endpoint group); a wrong value fails deterministically — don't retry.
• FROM graph is mandatory and differs per DB: ChEBI/Reactome → http://rdf.ebi.ac.uk/dataset/{chebi,reactome}; GO → http://rdfportal.org/ontology/go. Rhea and UniProt take no FROM. Wrong/missing graph → empty results, no error.
• Reactome string xref needs ^^xsd:string (bp:db "UniProt"^^xsd:string); a plain literal silently matches nothing.
• UniProt: always up:reviewed 1; never put bif:contains on a property path (split it). Rhea/Reactome/GO: always LIMIT on exploratory queries.
• Keyword search ≠ validation. The whole point of the workflow: every ID from a search_* call must be fed back into a SPARQL query, and you cite the SPARQL result, not the search snippet.

Troubleshooting

Symptom	Fix
ChEBI structure query returns 0 rows	Confirm chemrof predicates with SELECT ?p ?o { obo:CHEBI_16359 ?p ?o FILTER(isLiteral(?o)) }; namespace may have changed again
UniProt query ReadTimeout	You joined recommendedName with classifiedWith — split into two queries (see above)
run_sparql returns empty unexpectedly	Missing/wrong FROM graph, dropped ^^xsd:string, or chebi: instead of chemrof: — diff against the verified queries here
go: definitions returns 0	GO uses the primary endpoint and FROM <http://rdfportal.org/ontology/go> — not the ebi graph

Full method

references/research_article_analysis.md — the full phased checklist, ID-tracking templates, scoring rubric, and anti-skipping safeguards. This SKILL.md gives the corrected operational queries; that file gives the discipline and the deliverable structure.