Abstract Robust and trustworthy biomedical question answering (QA) remains a critical challenge for large language models (LLMs), especially in complex domains such as rare diseases, where information is fragmented across multiple sources. MedHopQA 2025 benchmark introduces 10 000 multi-step reasoning questions curated from Wikipedia, requiring systems to extract, connect, and synthesize biomedical knowledge across interlinked documents. In this work, we present a retrieval-augmented generation (RAG) and decision-making framework that integrates diverse retrieval strategies, including Query2Doc-based, Rationale-based, and Web-augmented retrieval, and employs a dedicated Decision Maker model to select or directly generate the most accurate and well-reasoned answers. Our system not only leverages evidence from both Wikipedia and the web but also explicitly evaluates and compares candidate answers to ensure answer reliability and reasoning transparency. Experiments on the test set demonstrate that our ensemble approach achieves state-of-the-art performance, yielding an Exact Match score of 0.84, and highlighting the importance of hybrid retrieval and robust decision-making in advancing biomedical multi-step QA. Furthermore, comprehensive analyses across retrieval stages reveal that mitigating model self-bias and ensuring a balanced integration of diverse sources are crucial for maximizing overall system performance, which ultimately reaches a score of 0.87.
Abstract Research aimed to valorize biodiversity, such as studies based on phytochemical characterization and bioactivity assessment, produces wide and heterogeneous datasets in which data collection is often non-standardized. This paper presents the design and implementation of a web-based plant metabolite database platform specifically developed to support biodiversity valorization research within the activities of the National Biodiversity Future Center, with a focus on the analysis of the Italian flora. The platform was designed to manage plant species, biological samples, experimental metabolites, molecular information, bioactivity data, and spectrometry results within an integrated and structured data environment. The system development followed an advanced, agile, and process-oriented methodology with strong and continuous involvement of biologists (i.e., the final users), from the collection and analysis of process- and data-related requirements to database modeling, considering the integration of existing external public databases. Following this initial phase, the development of the user interface proceeded, enabling researchers to insert data, perform advanced queries, and analyze and visualize spectral data through a dedicated tool. The web application was evaluated using the Questionnaire for User Interaction Satisfaction, which received high ratings from domain experts. This work demonstrates how the integration of process modeling and database design can enhance the reproducibility and standardization of research aimed at the valorization of biodiversity.
Type I interferon (IFN-I) is an important class of cytokines that can inhibit tumuor progression through mechanisms such as immunomodulation of the tumuor microenvironment or targeting cellular components. Although various endogenous and exogenous IFN-I therapeutic strategies have been developed, reports of immune cell dysfunction resulting from IFN-I signal enhancement indicate that optimal strategies have yet to be established. However, heterogeneous data of IFN-I are currently spread across multiple public databases and lack systematic integration, which poses challenges for knowledge acquisition and clinical research promotion. Herein, we develop the IFN-I Knowledge Base (IFNIKB), the first specific database for IFN-I. It integrates 26 273 literature articles on IFN-I antitumuor immunity, 2202 clinical trial records, and data on 8372 genes and 7164 proteins across 654 species. Furthermore, we design an automated workflow for knowledge discovery from literature. Users can construct on-demand knowledge graphs to explore entities and relationships through interactive visualizations, temporal trends, and network topology analyses. Additionally, we provide tools for multiple sequence alignment, sequence identity computation, and phylogenetic analysis to interpret IFN-I from molecular perspectives. Therefore, researchers can employ IFNIKB to conveniently acquire knowledge, propose hypothesis, optimize experimental design, and identify potential clinical therapeutic targets. Database URL: http://www.combio-lezhang.online/IFNIKB/home.
Members of the International Nucleotide Sequence Database Collaboration (INSDC; https://www.insdc.org/) collect, exchange, and preserve comprehensive open nucleotide sequence information and provide tools for its access. The INSDC has stated its commitment to welcoming new members into the collaboration to be more representative of the global community of data and users. To reach this goal, a comprehensive definition of the INSDC data model and minimum requirements for data acceptance have been established. Here we describe the processes used to arrive upon these INSDC Specifications and lay out strategies for their continued upkeep to remain current and relevant. Database URL: https://www.insdc.org/.
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.
