TogoMCP: natural language querying of life-science knowledge graphs via schema-guided LLMs and the Model Context Protocol
Database · 2026
DOI: 10.1093/database/baag042 ↗Kurz gefasst
Studienprotokoll in der Zeitschrift Database, 2026. Verfasst von Akira R Kinjo und 4 weiteren Autor:innen. Beteiligte Einrichtung: Anima Machina G.K., Osaka Station Building No. 3, 29th Floor, Room 1-1-1, 1-1-3 Umeda, Kita-ku, Osaka 530-0001, Japan. Hauptthemen laut MEDLINE: Databases, Factual, Software, Data Mining. Die Arbeit nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: MEXT, NBDC, Japan Science and Technology Agency.
Abstract (Originalsprache)
Querying the RDF Portal knowledge graph maintained by DBCLS—which aggregates ~60 life-science databases—requires proficiency in both SPARQL and database-specific RDF schemas, placing this resource beyond the reach of most researchers. Large Language Models (LLMs) can, in principle, translate natural-language questions into executable SPARQL, but without schema-level context, they frequently fabricate non-existent predicates or fail to resolve entity names to database-specific identifiers. We present TogoMCP, a system that recasts the LLM as a protocol-driven inference engine orchestrating specialized tools via the Model Context Protocol (MCP). Two mechanisms are essential to its design: (i) the MIE (Metadata-Interoperability-Exchange) file, a concise YAML document that dynamically supplies the LLM with each target database’s structural and semantic context at query time; and (ii) a two-stage workflow separating entity resolution via external REST APIs from schema-guided SPARQL generation. On a benchmark of 50 biologically grounded questions spanning five types and 23 databases, TogoMCP achieved a large improvement over an unaided baseline (Cohen’s $d = 1.82$, Wilcoxon $P \lt .001$), with win rates exceeding 80% for question types with precise, verifiable answers. An ablation study shows that all component configurations deliver significant improvements, with MIE schema files providing the largest marginal contribution on mean per-question score ($\Delta = +0.50$ relative to a no-MIE condition, two-sided Wilcoxon $P = .067$; 90% bootstrap CI $[+0.04,\,\,+0.94]$ excludes zero); a one-line instruction to load the relevant MIE file recovers the same mean improvement as a full procedural protocol, while the protocol additionally reduces downside risk (loss rate 1.6% vs. 4.8%, Fisher $P = .036$). These results suggest a general design principle: concise, dynamically delivered schema context is more valuable than complex orchestration logic for mean score performance, while procedural guidance plays a complementary role in narrowing variance. Database URL: https://togomcp.rdfportal.org/.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Autor:innen und Institutionen
- Anima Machina G.K., Osaka Station Building No. 3, 29th Floor, Room 1-1-1, 1-1-3 Umeda, Kita-ku, Osaka 530-0001, Japan.
- Database Center for Life Science, Joint Support-Center for Data Science Research, Research Organization of Information and Systems, 10-3 Midori-cho, Tachikawa, 190-0014 Tokyo, Japan.
- University of Oviedo, Department of Computer Science, C/ Federico García Lorca, S/N, Oviedo, Aturias, Spain.
- Bioinformation and DDBJ Center, National Institute of Genetics, Research Organization of Information and Systems, 1111 Yata, Mishima, 411-8540 Shizuoka, Japan.
ORCID-Profile: Akira R Kinjo, Yasunori Yamamoto, Samuel Bustamante-Larriet, Jose-Emilio Labra-Gayo, Takatomo Fujisawa
Förderung
- MEXT
- NBDC
- Japan Science and Technology Agency
Zitieren
APA
Kinjo, A. R., Yamamoto, Y., Bustamante-Larriet, S., Labra-Gayo, J. E., & Fujisawa, T. (2026). TogoMCP: natural language querying of life-science knowledge graphs via schema-guided LLMs and the Model Context Protocol. Database. https://doi.org/10.1093/database/baag042
Vancouver
Kinjo AR, Yamamoto Y, Bustamante-Larriet S, Labra-Gayo JE, Fujisawa T. TogoMCP: natural language querying of life-science knowledge graphs via schema-guided LLMs and the Model Context Protocol. Database. 2026. doi:10.1093/database/baag042
BibTeX
@article{Kinjo2026,
author = {Kinjo, Akira R and Yamamoto, Yasunori and Bustamante-Larriet, Samuel and Labra-Gayo, Jose-Emilio and Fujisawa, Takatomo},
title = {TogoMCP: natural language querying of life-science knowledge graphs via schema-guided LLMs and the Model Context Protocol},
journal = {Database},
year = {2026},
doi = {10.1093/database/baag042}
}
EUVIMED