EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

MedHopper: an agentic RAG-LLM system for multi-hop biomedical QA

Rustam Ruslanovich Taktashov iD, Nadezhda Yurievna Biziukova iD, Alexander Viktorovich Dmitriev iD, Olga Aleksandrovna Tarasova iD

Database · 15.01.2026 · 2026

DOI: 10.1093/database/baag047 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Database, online erschienen am 13.08.2026. Verfasst von Rustam Ruslanovich Taktashov und 3 weiteren Autor:innen. Beteiligte Einrichtung: Institute of Biomedical Chemistry, 10 bld. 8, Pogodinskaya str., 119121 Moscow, Russia. Hauptthemen laut MEDLINE: Software. Die Arbeit nennt 43 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Program for Basic Research in the Russian Federation.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This multi-hop biomedical QA system combines dense retrieval with a bounded agentic workflow for the MedHopQA (BioCreative IX) task, which requires short, entity-centric answers from a Wikipedia-derived corpus evaluated via exact match with normalization. Our approach replaces prior prompt decomposition with a controlled state machine that routes questions into four strategies (direct, definition, intersection, and multi-hop). The pipeline performs dense retrieval with cross-encoder reranking, executes up to three hops of intermediate entity extraction when needed, and applies answer validation with a bounded query-repair loop. On the MedHopQA test set (N = 1000), MedHopper achieves 0.55 Exact Match (MedHopQA evaluation metric) under official CodaBench evaluation. Ablation confirms capped multi-hop execution, reranking, and query repair as primary contributors; validation modules provide smaller consistent gains. Analysis reveals persistent challenges under exact-match scoring, including answer-type inconsistency (chromosome vs cytoband; yes/no vs symptom) and surface variation across plausible lexicalizations. All experiments were run on a single consumer GPU (RTX 5080) with no API dependencies, producing deterministic outputs at zero marginal cost per query.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

SoftwareBiocuration

Autor:innen und Institutionen

  • Institute of Biomedical Chemistry, 10 bld. 8, Pogodinskaya str., 119121 Moscow, Russia.

ORCID-Profile: Rustam Ruslanovich Taktashov, Nadezhda Yurievna Biziukova, Alexander Viktorovich Dmitriev, Olga Aleksandrovna Tarasova

Förderung

  • Program for Basic Research in the Russian Federation

Zitieren

APA

Taktashov, R. R., Biziukova, N. Y., Dmitriev, A. V., & Tarasova, O. A. (2026). MedHopper: an agentic RAG-LLM system for multi-hop biomedical QA. Database, 2026. https://doi.org/10.1093/database/baag047

Vancouver

Taktashov RR, Biziukova NY, Dmitriev AV, Tarasova OA. MedHopper: an agentic RAG-LLM system for multi-hop biomedical QA. Database. 2026;2026. doi:10.1093/database/baag047

BibTeX
@article{Taktashov2026,
  author = {Taktashov, Rustam Ruslanovich and Biziukova, Nadezhda Yurievna and Dmitriev, Alexander Viktorovich and Tarasova, Olga Aleksandrovna},
  title = {MedHopper: an agentic RAG-LLM system for multi-hop biomedical QA},
  journal = {Database},
  year = {2026},
  volume = {2026},
  doi = {10.1093/database/baag047}
}
Kostenlos starten →