Accepted submissions

The Keteninformatie Kerngegevens Verbeteren (KIK-V) program is a strategic initiative by Zorginstituut Nederland (Dutch National Health Care Institute) which aims to standardize and streamline the exchange of quality information in the nursing home sector, thereby reducing administrative burdens and enhancing data exchange for everything from Health Inspector visits to sharing personnel or information about capacity and capabilities. There are currently 15 different institutions participating in the programme, each with information exchange profiles consisting of tens of queries and functional specifications.

Motivation

Large health indicator programs (hundreds of measures and ‘validated questions’ across stakeholders) require SPARQL queries that are faithful to their Functional Descriptions (specifications) and internally consistent in naming, ontology usage, and parameter handling. In practice, query authors copy snippets, rename variables inconsistently, and diverge from the specified logic. Reviewers face a similar burden: manually aligning code with specifications and spotting divergence across near-duplicate indicators. The result is slow iteration, duplicated effort, and drift between the indicator specifications and the actual queries.

Why This Matters for Linked Data & Health

As steadily more health care institutions join the KIK-V network for information exchange, it is a challenge to schedule production of functional specifications and their corresponding queries with consistent terminology and policy alignment. KIKV-GraphRAG operationalizes linked data best practices: consistent vocabularies, reproducible queries, and transparent provenance, while scaling to hundreds of indicators. By unifying retrieval, generation, and review with context logging, it provides a repeatable path to high-quality SPARQL across health indicator portfolios.

Submission type

2. Demonstration (max 2 pages; use comment box for technical requirements)

Categories

C3 – SWAT4Health

T4 Semantic methods and AI

22

Embedding-based Deduplication of Knowledge Graphs using Graph Neural Networks

Emma Pinckers¹, Yulia Shapovalova², Shervin Mehryar¹, Michel Dumontier¹

¹Maastricht University, Maastricht, Netherlands. ²Radboud University, Nijmegen, Netherlands

Abstract

Knowledge graphs (KGs) built from multiple sources often contain duplicated entities caused by inconsistent naming, differing schemas, and incomplete updates, which reduce their reliability in applications such as research and decision making for life sciences and health care. Traditional deduplication approaches perform reasonably well on simple graphs but struggle to handle the scale and relational diversity of modern KGs. This paper explores how a Relation Graph Convolutional Network (R GCN) can overcome these limitations by learning from both the structure and semantics of heterogeneous relations. We train an R-GCN model and demonstrate its performance at various levels of scale and diversity in data. Through experimentation, we show that the proposed approach outperforms baseline models on both general purpose and clinical deduplication tasks. Over clinical datasets, the approach is further shown to be reliable and consistent using uncertainty quantification metrics.

Submission type

5. Long research paper (min 10 pages)

Categories

C2 – Tools and methods based on Semantic Web, knowledge representation and AI

Revised upload for the proceedings

Download file

23

Semantic Interoperability at National Scale: The SPHN Federated Clinical Routine Dataset

Jan Armida¹, Vasundra Touré¹, Philip Krauss², Deepak Unni¹, Harald Witte¹, Davide Chiarugi¹, Andrea Brites Marto¹, Julia Mauer³, Thomas Geiger³, Henning Beywl⁴, Marc Daverat⁵, Xeni Deligianni^6,7, Dominique Furrer⁸, Mathias Gassner⁹, Matthias Joos¹⁰, Katie Kalt¹⁰, Janshah Veettuvalappil Ikbal⁸, Helena Peic Tukuljac¹¹, Gaëlle Vuaridel-Thurre¹², Solange Zoergiebel¹², Sabine Österle¹

C4 – Data and models

Revised upload for the proceedings

Download file

47

Clinical Data Goes MEDS? Let’s OWL make sense of it

Alberto Marfoglia^1,2, Jong Ho Jhee², Adrien Coulet²

¹Dept. of Computer Science and Engineering – DISI, University of Bologna, Bologna, Italy. ²Inria, Inserm, Université Paris Cité, HeKA, UMR 1346, Paris, France

Abstract

The application of machine learning on healthcare data is often hindered by the lack of standardized and semantically explicit representation, leading to limited interoperability and reproducibility across datasets and experiments. The Medical Event Data Standard (MEDS) addresses these issues by introducing a minimal, event-centric data model designed for reproducible machine-learning workflows from health data. However, MEDS is defined as a data-format specification and does not natively provide integration with the Semantic Web ecosystem. In this article, we introduce MEDS-OWL, a lightweight OWL ontology that provides formal concepts and relations to represent MEDS datasets as RDF graphs. Additionally, we implemented meds2rdf, a Python conversion library that transforms MEDS events into RDF graphs, ensuring conformance with the ontology. We evaluate the proposed approach on two datasets: a synthetic clinical cohort describing care pathways for ruptured intracranial aneurysms, and a real-world subset of MIMIC-IV. To assess semantic consistency, we performed a SHACL validation against the resulting knowledge graphs. The first release of MEDS-OWL comprises 13 classes, 10 object properties, 20 data properties, and 24 OWL axioms. Combined with meds2rdf, it enables data transformation into FAIR-aligned datasets, provenance-aware publishing, and interoperability of event-based clinical data. By bridging MEDS with the Semantic Web, this work contributes a reusable semantic layer for event-based clinical data and establishes a robust foundation for subsequent graph-based analytics.

Submission type

5. Long research paper (min 10 pages)

Categories

C3 – SWAT4Health

Revised upload for the proceedings

Download file

52

Efficient Querying of Federated Large-Scale Clinical RDF Knowledge Graphs in the Swiss Personalized Health Network

Andrea Brites Marto¹, Philip Krauss², Katie Kalt³, Vasundra Touré¹, Deepak Unni¹, Sabine Österle¹

¹Swiss Personalized Health Network, Basel, Switzerland. ²Accenture AG, Basel, Switzerland. ³University Hospital of Zurich, Zurich, Switzerland

Abstract

The Swiss Personalized Health Network developed a national federated framework for semantically
described medical data, in particular hospital clinical routine data. Instead of centralizing patient-level
information, hospitals perform semantic coding and standardization locally and store SPHN-compliant data
in a triple store. These decentralized RDF datasets, following the FAIR (Findable, Accessible, Interoperable, Reusable) principles, together exceed 12 billion triples across more than 800,000 patients, all signed a broad consent.
In this work, we address the computational challenge of efficiently querying and integrating these
distributed RDF resources through SPARQL. Our use cases focus on feasibility queries and value distribution, which allow researchers to assess the potential availability of patient cohorts across hospitals without disclosing sensitive patient-level information. We present methods for optimizing SPARQL querying, tailored to the characteristics of large-scale federated and complex clinical data.
We evaluate these approaches by iteratively testing optimized queries on the SPHN Federated Clinical
Routine Dataset, which spans 125 SPHN concepts including demographics, diagnoses, procedures,
medications, laboratory results, vital signs, clinical scores, allergies, microbiology, intensive care data,
oncology, and biological samples. With this approach, we’ve built a set of rules to consider for gradually
optimizing SPARQL queries. Our results demonstrate that optimized SPARQL query planning and execution can significantly reduce response times without compromising semantic interoperability.

Submission type

5. Long research paper (min 10 pages)

Categories

D1 – RDF Dataset or SPARQL endpoint

Revised upload for the proceedings

Download file

62

Modular composition of SPARQL queries for focusing onwhat to look for rather than how to get it

Yael Tirlet¹, Jerven Bolleman², Emmanuelle Becker¹, Fabrice Legeai³, Olivier Dameron¹

¹Univ Rennes, Inria, CNRS, IRISA – UMR 6074, F-35000 Rennes, France, Rennes, France. ²SIB Swiss Institute of Bioinformatics, 1, rue Michel Servet – CH 1211 Geneva 4 – Switzerland, Geneva, Switzerland. ³IGEPP, INRAE, Institut Agro, Univ Rennes, 35653, Le Rheu, France, Le Rheu, France

Abstract

Adoption of life science knowledge bases by domain experts remains low in spite of the increasing accessibility of these bases, as the Semantic Web framework supports advanced integration and querying. The main bottleneck for leveraging these knowledge bases is that advanced querying combines the inner complexity of life sciences (which requires domain expertise) with the technical complexity of knowledge bases schemas and of SPARQL (which requires engineering skills).

We propose a framework based on modules that reconciles both views. A module corresponds to a concept relevant to domain experts and is associated with a SPARQL fragment compliant with the data schema. Modules can be connected to compose new modules corresponding to more complex concepts; the SPARQL fragments of the components are automatically combined to constitute the fragment of the composed module. Our approach thus allows experts to focus on what they are looking for, while our system takes care of how to obtain it.

Submission type

5. Long research paper (min 10 pages)

Categories

C2 – Tools and methods based on Semantic Web, knowledge representation and AI

78

Usage of mapping properties on the semantic web to align the existing variety in the semantics of Lyme disease

Pauline Lubet¹, Daniel Fernández-Álvarez², Ronald Cornet¹, Andra Waagmeester¹

¹Department of Medical Informatics, Reusable Health Data group Amsterdam Public Health Research Institute,, Amsterdam, Netherlands. ²WESO Research Group, University of Oviedo, Oviedo, Spain

Abstract

Disease entities are connected across biomedical semantic resources through a variety of mapping relations, whose semantics and usage differ between data providers. As a result, a single disease is not described in one place in the biomedical semantic web but across multiple resources, each providing partial and complementary representations. These differences influence how disease knowledge is interpreted when representations from multiple sources are combined.

In this paper, we analyse how mapping choices shape disease-level integration by examining the representation of Lyme disease across RDF-based biomedical resources. Starting from the Orphanet Rare Disease Ontology (ORDO), we iteratively follow inter-resource links to identify disease entities and mappings across 17 connected resources, eight of which are openly available for quantitative analysis.

Our results show that inter-resource links rely predominantly on cross-reference properties such as oboInOwl:hasDbXref, while SKOS mapping relations are used more selectively and unevenly across resources. At the disease level, we observe that entities corresponding to different clinical or modelling scopes are sometimes connected using equivalence-oriented relations, with implications for how distinctions preserved within individual resources are reflected in integrated representations.

To support the interpretation of these complex mapping structures and workflow, we develop a first visualization prototype that provides an integrated view of disease entities, mapping properties, and resource coverage. Overall, the study illustrates how mapping practices influence integrated disease representations and underlines the importance of making mapping semantics explicit for FAIR reuse of biomedical knowledge.

Submission type

5. Long research paper (min 10 pages)

Categories

C4 – Data and models

Accepted Posters

24

From VCF to RDF: RML-Based Conversion Approaches for the Semantic Representation of Variant Data

Elias Crum^1,2, Bart Buelens², Gökhan Ertaylan², Ruben Taelman¹

¹Ghent University, Gent, Belgium. ²VITO NV, Mol, Belgium

Abstract

Abstract

Lifestyle interventions can reduce the impact of chronic diseases such as type 2 diabetes, prediabetes, and cardiovascular disease. The process of generating personalized lifestyle advice is largely manual and time-consuming, as it requires comparing individual biomedical and lifestyle data with evidence-based recommendations.

A semantic model is proposed to support the automated generation of personalised lifestyle advice. The model integrates biomedical measurements, lifestyle behaviour, and interpretation protocols in a structured framework, enabling explicit linking between data, interpretations, and lifestyle change plans. In combination with appropriate tools, this model provides a foundation for generating scientifically grounded lifestyle advice based on an individual’s biomedical and lifestyle data.

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

C4 – Data and models

40

SHAIPED Deliverable 3.7: Demonstrator showcasing data access to augment EHDS dataset catalogues with Generative AI

Derycke Pascal

Sciensano, Bruxelles, Belgium

Abstract

Building upon the HealthDCAT-AP RDF vocabulary, the SHAIPED D3.7 demonstrator showcases an augmented dataset catalogue that leverages Generative AI (GenAI) to enhance dataset discovery. Within the context of the European Health Data Space (EHDS), the demonstrator translates research questions into semantically coherent dataset representations through a Virtual Dataset Builder. Using retrieval-augmented generation (RAG) and large language model embeddings, the system identifies relevant dataset variables, constructs conceptual “virtual datasets,” and proposes synthetic proxy data to support privacy-preserving exploration and early-stage prototyping analyses. This approach provides a human-friendly layer on top of semantic web technologies such as DCAT, CSVW, and Wikidata, bridging the gap between sensitive health data and research applications.

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

C2 – Tools and methods based on Semantic Web, knowledge representation and AI

44

RO-Crates for BioImaging

Tiago Lubiana¹, Susanne Kunis², Josh Moore¹

¹German BioImaging-Society for Microscopy and Image Analysis e.V., Constance, Germany. ²enter of Cellular Nanoanalytics, Integrated Bioimaging Facility iBiOs, University of Osnabrück, Osnabrück, Germany

Abstract

Bioimaging research increasingly relies on complex, high-volume microscopy data that is still rarely shared in a FAIR and interoperable way. While formats such as OME-Zarr address modern, cloud-native data storage and access, the bioimaging community continues to lack Linked Data–ready metadata standards that capture experimental context and relationships between research artefacts. Research Object Crates (RO-Crates) provide a promising, JSON-LD–based mechanism to package data together with rich, standardized metadata. In this work, we explore how RO-Crates can complement existing Open Microscopy Environment standards, with a focus on OME-Zarr. We present recent practical experiences, including the use of RO-Crates in the OME 2024 NGFF Challenge, and discuss ongoing efforts toward formal RO-Crate profiles and metadata pipelines for major bioimaging repositories. Together, these efforts highlight RO-Crate as a valuable addition to the modern bioimaging metadata toolkit.

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

C4 – Data and models

Revised upload for the proceedings

Download file

46

Making research resources in Life Sciences and Health interoperable

Walter Baccinelli, Vedran Kasalica

Health-RI, Utrecht, Netherlands

Abstract

Interoperability challenges continue to limit the adoption of FAIR data practices across the Dutch Life Sciences and Health (LSH) domain. To address this, a four-year project within the Thematic Digital Competence Center – Life Sciences and Health (TDCC-LSH), started in June 2025. The project combines community-driven alignment with technical coordination to support interoperable research data workflows. Key activities include semantic interoperability using an upper-level ontology, harmonisation of research tools through ontologies (e.g., EDAM), and collaboration with ELIXIR, Research Data Alliance working groups, and large-scale research infrastructures.

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

T3 FAIR4HCLS

50

AI-augmented curation of drug vocabulary for knowledge-graph–driven repurposing and analytics

Anneli Karlsson, Rebecca Foulger, Mark Streer, Brandon Walts, Joseph Mullen, Mark McDowall

SciBite from Elsevier, BioData Innovation Centre, Wellcome Genome Campus Hinxton, Cambridge, United Kingdom

Abstract

Abstract

Abstract

Managing and integrating metagenomic data is a key challenge in holobiont studies, as understanding host–microbiota interactions requires linking complex heterogeneous datasets, such as microbial diversity, host phenotype, and environmental context. In response, Holograph, an RDF schema dedicated to the representation of holobiont data, has been developed to provide a structured and interoperable way to store such heterogeneous information. It includes a central part dedicated to handle metabarcoding data, i.e. abundance tables of Amplicon Sequence Variants, which is connected to various descriptors, called features of interest, of the host itself or its environment. Observations and variables related to the host are handled with an implementation of the SOSA and I-ADOPT ontologies. In addition, the GeoSPARQL ontology is used to define complex spatial relationships between the locations of features of interest such as the biological compartment where the sample has been extracted, the corresponding host (e.g plant) and geographical information (e.g. plot where the plant is cultivated and the field containing the plot). As a result, this generic schema was applied to integrate the data from a large case study covering metagenomics, biochemical assays, bioagressors description, climatic and phenotypic data. Furthermore, the Holograph’s schema is also currently implemented on livestock holobionts datasets and can be queried with a dedicated web-interface.

Keywords: Holobiont, Metagenomics, Agroecological, RDF, Ontology

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

C4 – Data and models

Revised upload for the proceedings

Download file

70

FAIR-Mind: Application Profile for Depression Clinical Assessments and HRV Biomarker Integration

Inkyung Choi¹, Milena Čukić²

¹Sungkyunkwan University, Seoul, Korea, Republic of. ²OST University of Applied Sciences of Eastern Switzerland, St. Gallen, Switzerland

Abstract

Mental health research increasingly combines clinical assessments with physiological biomarkers, yet semantic interoperability remains limited compared to oncology and neuroimaging. FAIR-Mind is a draft Application Profile (AP) integrating depression clinical data with ECG-derived heart rate variability (HRV) biomarkers. Based on UK Biobank data structure and recent findings linking HRV to suicide risk (Weber et al., 2025), we mapped 22 variables across depression assessments (PHQ-9, CIDI, suicidality) and HRV metrics to 8 ontologies: CMO, BMONT, IOBC, BCIO, PROV-O, NCIT, CHEBI, and SNOMED CT. Ten competency questions spanning data discovery, integration, provenance, and clinical interpretation guide AP design and validation. This work demonstrates practical AP methodology for mental health research addressing calls for psychiatric ontology development (McInnis et al., 2025). The draft AP awaits UK Biobank data access (Q1 2026) for runtime validation, with community input sought on three open design questions.

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

T3 FAIR4HCLS

72

iMapper: An Interactive Tool for Biomedical Concept Harmonization and Semantic Mapping

Sathvik Guru Rao, Alpha Tom Kodamullil

Fraunhofer SCAI, Sankt Augustin, Germany

Abstract

Data curation, mapping, and harmonization are foundational activities in the data lifecycle,

forming the cornerstone of effective data stewardship and enabling the reuse, interoperability, and

reproducibility of research data. In biomedical and clinical research, these activities ensure that data

adhere to the FAIR (Findable, Accessible, Interoperable, and Reusable) principles by creating explicit

semantic relationships between heterogeneous data sources. Recent developments in large

language models (LLMs) and retrieval-augmented generation (RAG) offer new opportunities to

transform this landscape providing efficient automated suggestions but they often focus on exact

matches and lack mechanisms for reasoning or handling imperfect but valuable correspondences. In

practice, human expertise remains critical, as curators must validate mappings, resolve ambiguities,

and ensure semantic accuracy.

To address this need, we present an interactive biomedical concept harmonization and semantic

mapping tool that combines Retrieval-Augmented Generation (RAG) based mapping with LLM-

based reasoning. The system retrieves candidate matches for input entities, infers relationships

Andra Waagmeester¹, Machteld Boonstra¹, Sebastian van der Voort¹, Lydia Pintscher², Sabine Oesterle³

¹Department of Medical Informatics, Reusable Health Data group, Amsterdam Public Health Research Institute, Methodology Digital Health, Location AMC Meibergdreef 9, 1105 AZ Amsterdam, Amsterdam, Netherlands. ²lydia.pintscher@wikimedia.de, Berlin, Germany. ³SIB Swiss Institute of Bioinformatics, Basel, Switzerland, Basel, Switzerland

Abstract

The adjective “federated” is widely used in computer science, yet with diverging meanings that often lead to confusion, especially in interdisciplinary discussions, collaborative authorship or when building research consortia. Originally rooted in political theory, federation denotes cooperation between autonomous entities without loss of sovereignty. Recently, however, “federated” now also describes architectures ranging from distributed querying and data integration to machine learning, research infrastructures, and collaborative knowledge or data ecosystems.

Motivated by recurring ambiguity in discussions among co-authors, this poster tries to distinguish prominent uses of the adjective “federated”. (e.g. federated SPARQL querying, federated learning, federated data warehousing, national initiatives such as the Swiss Personalized Health Network (SPHN), and the Wikimedia ecosystem). We review where these concepts overlap, where they diverge, and whether they should be treated as distinct notions. By making implicit assumptions explicit, we try to bring some clarity and hopefully practical guidance on when and how the term “ federated” should be used.

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

A1 – Abstract submission – Thematic

85

Towards HealthDCAT-AP compliant provenance description of clinical trial data flows according to ICH E6

Matthias Löbe¹, Judith Wodke²

¹IMISE Universität, Leipzig, Germany. ²UCM Universitätsmedizin, Greifswald, Germany

Abstract

Many researchers consider the provenance of data sets to be very important when it comes to trust and reusability of medical data sets. While the topic is being intensively studied in academia, information on data origin is often lacking in practice. One reason for this is the lack of coordinated, easily applicable recommendations for guidelines to produce provenance statements that can be interpreted comparatively across data sets. The HealthDCAT-AP vocabulary was developed to describe data sets in national catalogs and enables simple as well as complex provenance statements using the W3C PROV model. In this article, the data flows of an exemplary clinical trial were implemented. Since the adoption of Revision 3 of the ICH E6 guideline, data flow diagrams should be included as part of sponsors’ data management plans. The results show that data flows can be generated and visualized straightforwardly with PROV. However, there is a lack of coordinated vocabularies for the types of actors, activities, and entities that occur in such data flows to make data flow diagrams easy to read across studies and to enable queries, for example, on the characteristics of data collection.

Submission type

1. Poster (max 2 pages, including reports on applications, data, and models)

Categories

C2 – Tools and methods based on Semantic Web, knowledge representation and AI

90

Using FHIR and Wikidata to exchange neighborhood geo-referenced contextual data to personalize walking interventions

Xi YANG^1,2, Pauline Lubet¹, Daniel Fernández-Álvarez³, Andra Waagmeester¹, Martin Dijst^2,4, Ronald Cornet¹

C4 – Data and models

26