With the rapid advances in digitalization and data-driven biomedical research, the volume and complexity of health-related datasets have grown significantly in recent years. These datasets, such as those within the Swiss Personalized Health Network (SPHN) [1], are often collected across multiple institutions and hold immense potential for advancing personalized medicine and clinical decision-making. Preparing these datasets for research use requires substantial time and effort for the curation, standardization, and cleaning of data.

Despite the effort to generate them, curated datasets are rarely reused outside their original consortia. One seemingly trivial reason is unawareness about suitable datasets due to their limited discoverability. Another reason is the legal and regulatory access restrictions to clinical datasets. Such datasets often comprise sensitive patient-related data and therefore cannot simply be shared with or previewed to third parties potentially interested in reuse. The provision of rich, findable, and easily accessible metadata, that is, data about data, would be a workaround to these key barriers by improving insights into the contents of a dataset without revealing sensitive information. Unfortunately, structured and accessible metadata that adequately describes the dataset content is often lacking. In many cases, available metadata is limited to high-level descriptive information, which provides insufficient insight into the structure, semantics, and actual content of the datasets. As a result, researchers frequently struggle to identify relevant datasets or, due to incomplete metadata, are unable to make a well-informed decision if the dataset is useful for their specific research questions. This ultimately hinders the broader reuse of these valuable data resources.

The Findable, Accessible, Interoperable, and Reusable (FAIR) data principles, formally introduced in 2016 [2], establish guidelines to ensure that data can be discoverable, accessible, and reusable by both humans and machines. While these principles have gained widespread adoption, the FAIR principles build on a lineage of initiatives aimed at improving metadata standards and data stewardship. Foundational efforts such as the Dublin Core Metadata Initiative (DCMI) [3], the Data Documentation Initiative (DDI) [4] in the social sciences, and the Open Archival Information System reference model (International Organization for Standardization 14721:2003) [5] established essential models for structuring metadata and long-term preservation. The Semantic Web vision [6] further advanced the idea of machine-readable, linked metadata to enhance data integration and discoverability.

Several well-established and widely adopted standards from the World Wide Web Consortium (W3C) [7] provide a standardized vocabulary to describe digital objects:

In addition to these general ones, there are also metadata standards for describing health-related datasets:

Building on this foundation of existing standards, the FDP specification [18] provides a standardized, layered architecture for exposing metadata in a FAIR-compliant, machine-actionable manner. An FDP [19] is a metadata service designed to support the publication and discovery of metadata according to the FAIR principles. It provides a Representational State Transfer API for the creation, storage, and retrieval of FAIR-compliant metadata. Serving a dual role enables data owners to expose metadata about their digital objects in a FAIR manner and at the same time facilitates metadata discovery by consumers seeking to access and use the resources.

Metadata catalogs for datasets play a critical role in enabling FAIR data principles [2]. Several well-established national and international metadata cataloging efforts exist, most of which leverage the aforementioned W3C standards to represent the metadata about datasets that are well structured and machine-readable. The following support the representation of metadata for health-related datasets:

The examples demonstrate that metadata catalogs improve dataset visibility and accessibility. However, much existing catalogs' metadata is not tailored to health-related datasets and remains limited to top-level dataset descriptions, providing little insight into the actual structure or semantics of the underlying data. A comparison of the characteristics of existing metadata catalogs is provided in Table 1.

In Switzerland, SPHN was launched in 2017 by the Swiss Federal Government to develop a sustainable and interoperable national data infrastructure for health-related research. SPHN promotes the sharing of clinical and omics data across institutions, supported by a semantic framework based on RDF and ontologies [31,32]. This framework consists of a set of data schemas: the SPHN RDF Schema as the foundation, along with related project-specific schemas that extend it. Together, these schemas define the semantics and constraints to which SPHN data must conform. To foster FAIR data practices within the network, SPHN encourages the publication of structured metadata describing datasets, thereby enhancing transparency and promoting the responsible reuse of data. A key enabler for effective secondary use of health data is the ability for researchers to identify, explore, and evaluate existing datasets before initiating new data collection efforts or collaborations.

To address this need, SPHN developed the SPHN Metadata Catalog [33], a national infrastructure that provides centralized access to semantically rich metadata for health-related datasets. While primarily focused on datasets generated by SPHN-funded projects, the catalog is designed to scale and support metadata from other relevant and related initiatives in Switzerland. The SPHN Metadata Catalog offers HealthDCAT-AP-compliant metadata, an FDP (including a public SPARQL Protocol and RDF Query Language [SPARQL] endpoint), aggregated instance counts of the knowledge graphs underlying the metadata (ie, quantitative metadata), as well as metadata visualization features [34]. These features allow users to inspect and interactively explore dataset content, such as variables and their relations, data types and coding systems used, and summary statistics that reflect the abundance of specific data elements without exposing sensitive or individual-level data. This bridges the gap between high-level dataset overviews and technical data dictionaries and supports informed decision-making. In this paper, we describe the design and implementation of the SPHN Metadata Catalog, including the underlying metadata model, semantic technologies, the development process, and discuss its role in enabling FAIR and scalable metadata-driven data discovery for health research in Switzerland.

Ethics approval and Institutional Review Board (IRB) review were not required for this study because no identifiable patient-level information was accessed or shared.

Systematic metadata collection is critical to ensure that research projects and their datasets are discoverable, interoperable, and reusable. To achieve this, metadata is collected along 2 complementary dimensions:

One key part of the SPHN Metadata Catalog is the SPHN FDP [33], as illustrated in Figure 4, which extends the reference implementation of the FDP specification. It ingests structured metadata (in RDF), the output of the SPHN Metacato (Figure 1). While the reference implementation of the FDP provides a standardized way to expose metadata in both human and machine-readable manner, the SPHN FDP builds on this foundation to meet specific needs of SPHN. For example, the introduction of new metadata elements that are required to describe routine clinical datasets, reuse of existing metadata elements from HealthDCAT-AP, and modification of the UI to concisely display certain metadata. By extending the reference implementation, the SPHN FDP stays aligned with the FDP specification, ensuring compatibility with broader FAIR data initiatives and tools. The SPHN FDP is backed by a GraphDB triple store as the primary store for the metadata records.

The use of semantic standards for representing metadata ensures that metadata managed within the SPHN FDP is interoperable, machine-readable, and machine-actionable. This is critical for enabling automated discovery and reuse of health-related datasets. Since the SPHN FDP is based on the FDP reference implementation, it adopts a set of core FDP-specific standards that establish a baseline for FAIR compliance and interoperability. The DCMI Metadata Terms provide a general foundation for describing catalogs and datasets, DCAT enables datasets to be structured to support both human and machine discovery, and FDP-O extends DCAT with additional metadata elements required for managing metadata within an FDP. Building on this foundation, the SPHN FDP incorporates domain-specific standards such as HealthDCAT-AP to address the requirements of health-related metadata in Switzerland. To further complement these established standards, we introduce the SPHN Metadata Catalog Schema [37] (Figure 2). This lightweight schema follows the overall structure of DCAT and HealthDCAT-AP while accommodating SPHN-specific metadata elements that fall outside the scope of FDP-O. It is specifically designed for metadata representation within the SPHN FDP [33] as part of the SPHN Metadata Catalog.

The SPHN Metadata Catalog Schema (licensed under CC BY 4.0) consists of 2 parts: a minimal SPHN extension to DCAT and FDP-O that defines SPHN-specific properties and their semantics, and a set of Shapes Constraint Language (SHACL) shapes [40] to document and specify the structure, cardinality, and other restrictions associated with elements of the schema (see “Metadata Validation”). The shapes allow the validation of metadata in the SPHN Metadata Catalog and ensure that the records conform to the SPHN Metadata Catalog Schema. Figure 2 provides a conceptual overview of the 4 classes that form the backbone of metadata representation in the SPHN FDP. These classes are primarily based on the DCAT vocabulary (with additional use of VoID), and they structure how projects, datasets, and their technical representations are described in the SPHN Metadata Catalog.

To illustrate with an example, a project funded under SPHN may be modeled as a DCAT Catalog. Each of the project’s datasets (eg, routine clinical data, genomic data, imaging data) is represented as an individual DCAT Dataset entity of the DCAT Catalog. For each dataset, the technical formats in which the data are available (eg, RDF, CSV, and JSON) are specified in the DCAT Distribution. As a dataset may be available in multiple formats, the DCAT Distribution allows users to choose the most suitable option. Quantitative metadata about a specific distribution is finally captured in a VoID Dataset, for example, the number of subjects, classes, or triples in a graph, complemented with statistical or structural summaries useful for describing data.

Each of the 4 classes (ie, DCAT Catalog, DCAT Dataset, DCAT Distribution, and VoID Dataset) has properties from DCMI Metadata Terms, DCAT, DCAT-AP, and HealthDCAT-AP to represent resource-specific metadata in a way that is consistent and interoperable. In addition, the SPHN Metadata Catalog Schema introduces new properties for capturing SPHN-specific metadata at each level of representation to further characterize and describe resources. A complete overview of the metadata elements defined in the schema is provided in Figure 3 and in the online documentation [37].

Ensuring the quality and conformance of metadata records within the SPHN FDP requires a reliable validation mechanism. Validation is performed against the SHACL shapes defined as part of the SPHN Metadata Catalog Schema [41] which encode the structural and semantic constraints that each metadata record must satisfy before it can be accepted into the SPHN Metadata Catalog. The full set of SHACL shapes is available at reference [40]. The SHACL shapes capture a range of constraint types. Property cardinality rules specify which metadata elements are mandatory or optional, ensuring that records carry a minimum level of descriptive completeness. Datatype and range constraints enforce that literal values conform to expected formats. Controlled vocabulary constraints restrict certain properties to values from defined value sets or ontologies. Together, these constraints validate the requirements set out by the SPHN Metadata Catalog Schema and its upstream standards, including DCAT, DCAT-AP, and HealthDCAT-AP.

Validation is integrated into the SPHN Metacato. When a metadata record is created or updated, the record is evaluated against the applicable SHACL shapes prior to being loaded into the SPHN FDP. Records that fail validation are rejected, and the resulting SHACL validation report is used to identify and mitigate the specific constraint violations. In addition to validation at creation time, periodic revalidation of existing catalog records can be triggered to assess continued conformance as the schema evolves. When the SPHN Metadata Catalog Schema is updated, existing records can be re-evaluated against the revised SHACL shapes. This supports the long-term maintainability of the catalog and ensures that the metadata records remain aligned with current schema versions.

The SPHN FDP exposes metadata through its UI, allowing users to explore descriptive information about datasets and resources in a structured way. To enhance interpretability, the metadata elements displayed in the UI are not static text. Wherever possible, following the linked data principles [42], their values are presented as hyperlinks that resolve to additional resources, either within the FDP or to external portals, vocabularies, and ontologies. For example, a dataset’s declared license might be a clickable link pointing to its formal definition, or associated publications that link out to further description about the publication itself. This design ensures that users can not only read metadata but also navigate across related resources for a comprehensive view about a catalog or dataset. Beyond text-based metadata, the SPHN FDP also provides references to interactive visualization with links directly to SPHN Schema Scope (described below) in the field “Interactive Exploration” that help users contextualize datasets and their schemas. When available, users can also explore the quantitative statistics associated with a dataset via SPHN Schema Scope to get a better sense of the composition of the dataset.

The second key component of the SPHN Metadata Catalog is the SPHN Schema Scope [34], an interactive exploration tool for SPHN(-related) project schemas (Figure 1). The SPHN Schema Scope is openly available under an open-source GPLv3 license [43]. It ingests SPHN Datasets (tabular representation of the SPHN Schema; [44]) and quality-controlled quantitative metadata in the form of an SQLite database (output of the SPHN Metadata Generator; Figure 1) to generate an interactively explorable schema graph. With its intuitive graphical user interface (GUI) and the accessible visual display of data schemas, it provides an entry point to a variety of user groups, catering to a wide range of background knowledge on graph-based data models. By following the connections in the graph, users can visually navigate classes, relationships, and hierarchies defined in the schema, making it easier to understand the semantic structure of the data. The SPHN Schema Scope is aligned and cross-linked with the SPHN FDP, allowing seamless transition and access to the SPHN Metadata Catalog.

The GUI of the SPHN Schema Scope application is based on the Shiny Web Application Framework for R (R Foundation for Statistical Computing) [45]. It is laid out to accommodate a broad range of user expertise, suitable for a quick overview for browsing to more in-depth exploration for potential secondary users. The GUI allows setting specifics of the graph representation and various general display attributes. The settings are passed to the server part of the application as parameters for graph rendering and GUI appearance. Where applicable, a tab “FAIR Data Point” is displayed, providing links to corresponding sites of the SPHN FDP [33]. User feedback has been incorporated to add functionalities considered important by the user community.

The graph visualization of SPHN Schema Scope is primarily based on the package visNetwork (Network Visualization using the “vis.js“ Library) [46]. To build the schema graph, classes, relationships, data types, and standards are first extracted from a structured tabular schema representation. For accessibility, that is, human readability, not only classes but also data types and terminologies are represented as nodes while relationships (predicates) are represented as directed edges of the schema graph. These 2 sets of nodes and edges are combined into a visNetwork-graph object, which is rendered by the server part of a Shiny application [45]. Graph display specifications can be adjusted through the GUI.

The GUI also allows the selection of a subset of concepts in the schema to accommodate for different areas of interest when exploring the graph. The depth of connections can be specified to explore elements not directly connected to the user’s selection. Likewise, the directionality can be set to assess the attributes of selected concepts or the elements referencing them. This helps to get a better understanding of the schema and the connectivity of its individual elements.

To derive the graph subset, an adjacency matrix of all nodes is derived in a first step from the set of nodes and edges previously determined (see section “Schema Visualization”). From this, a distance matrix is built using the igraph package (Network Analysis and Visualization) [47]. This distance matrix allows selecting the nodes in the scope of the selection and depth specified by the user. A restricted set of nodes is then used as input to derive the graph as described before.

For quantitative metadata display, the schema graph can be overlaid with count information from an SQLite database generated by the SPHN Metadata Generator (see section “Quantitative Metadata”). To this end, the Shiny application connects to this SQLite database, matches the table containing the data of interest with the corresponding data schema, and queries the database during runtime. The number of instances of the data elements is retrieved, and the count information is added as an attribute to the set of nodes and edges used for visualization. For classes, the count of distinct instances is provided. For terminologies, both the count of distinct codes and the total usage count are presented. For relationships, the number of occurrences in the data is shown. The size of nodes (classes and terminologies) is scaled according to the count information. A user can interactively set specifics of the display mode like the scaling mode via the GUI. Notably, the aggregated concept counts apply to the full graph only and do not allow any combinatorics or subcohort construction. Therefore, only the full graph can be explored when visualizing data frequencies to ensure correctness of the displayed data.

As components of the SPHN Metadata Catalog, both SPHN FDP [33] and SPHN Schema Scope [34] are deployed on virtual machines provided by SWITCH [48], a Swiss provider for secure cloud and networking for academic institutions, and are publicly available. The applications run inside Docker containers, ensuring portability and consistency across environments. Deployment-specific parameters such as environment variables (including database and schema specifications), storage volumes, general access modalities, port mappings, and container networking are defined in a Docker Compose YAML file that orchestrates the different services. A continuous integration/continuous deployment pipeline automates testing, building, and deployment from the GitLab-based repository to the virtual server infrastructure. Any change committed to the repository, for example, an update to the Docker Compose YAML or related configuration files, triggers a pipeline that redeploys the service. External access to the SPHN Metadata Catalog components is managed by Traefik (Traefik Labs) [49], a modern reverse proxy and load balancer tightly integrated with Docker. Traefik dynamically routes incoming internet traffic to the correct service and handles concerns such as transport layer security termination, secure routing, and traffic monitoring.

The SPHN Metadata Catalog was developed and deployed to enhance the findability and reusability of datasets generated within the SPHN. The system is in production and publicly accessible [33], and its implementation follows the FAIR principles, with a particular emphasis on providing semantically rich, machine-readable metadata to support dataset discovery. The architecture of the SPHN Metadata Catalog adheres to the FDP reference implementation [18], ensuring compliance with established standards for describing catalogs and datasets. The metadata is organized across the 3 levels—Catalog, Dataset, and Distribution—following the DCAT vocabulary. Each dataset entry is annotated with both generic and domain-specific metadata, aligning with established metadata conventions. In addition to DCAT (Figure 4), we used VoID vocabulary to represent dataset-level statistics to further describe the structure of the datasets. For SPHN-specific needs, the SPHN Metadata Catalog Schema enriches resources with health-related metadata such as inclusion and exclusion criteria, data modalities (eg, clinical, genomic, and imaging), patient age ranges, and gender distributions (see Figure 3 for the full SPHN Metadata Catalog Schema).

Metadata collection and management follow a structured and reproducible workflow. Initially, researchers provide dataset descriptions using a standardized Excel-based metadata template, structured according to the SPHN Metadata Catalog Schema. The filled templates are then processed through the SPHN Metacato that harmonizes and converts tabular entries of metadata into corresponding RDF representations. This conversion ensures syntactic and semantic consistency across datasets contributed by different projects by enforcing conformance with the underlying schema and produces metadata records compatible with the FDP infrastructure. The RDF metadata is subsequently ingested into the SPHN FDP instance, making it discoverable through both the web interface and API endpoints.

To serve the needs of diverse user groups, the SPHN Metadata Catalog provides both human-readable and machine-readable views of metadata (Figure 5 and Textbox 1, respectively). Typical users can browse and search datasets through a web-based interface, while automated agents and third-party systems can access the same information programmatically using the API. This dual-access approach supports discovery and evaluation of dataset relevance without exposing sensitive patient-level data.

To promote discovery and reuse in biomedical research, the SPHN Metadata Catalog is indexed and linked to the FDP registry [51], thereby integrating the SPHN FDP instance into the broader FAIR ecosystem. This connection demonstrates the interoperability of the SPHN Metadata Catalog with external FAIR-aligned data infrastructures and enables federated discovery. Through this integration, datasets published within the SPHN Metadata Catalog become part of a larger network of FAIR data resources.

Textbox 1 shows an excerpt of the metadata record presented in Figure 5 in RDF (Turtle syntax). Note that the order of elements varies in the RDF serialization. The full metadata record can be found in Multimedia Appendix 1 or online in the SPHN Metadata Catalog [50].

The SPHN Metadata Catalog is a recent addition to the resources provided by the SPHN and has been publicly available since May 2025. Despite its novelty, multiple projects and institutions have contributed content already. Statistics of website visits confirm a regular use of the SPHN Metadata Catalog, bringing improved visibility for projects and datasets represented there. Several projects have contributed more than one dataset, and for multiple datasets, more than one distribution is available (eg, corresponding data in different formats). The high number of metadata fields gives the opportunity to provide rich information. Contributors embrace this option, reflected by a high completion rate even for optional elements. For example, catalogs and datasets feature 84.2% and 72.9% completion rates across a total of 38 and 48 metadata fields, respectively. Table 2 summarizes the adoption metrics reached since the catalog’s launch.

The catalog advances all aspects of the FAIR principles, as it provides semantically rich, detailed, and accessible metadata that go well beyond the high-level descriptions typically found in classical metadata catalogs. The metadata is accessible to both humans and machines, supporting interoperability and reuse, with minor limitations remaining in free-text fields such as data access conditions or inclusion/exclusion criteria. Table 3 summarizes, for each principle, how the SPHN Metadata Catalog enables their support.

This paper presents the SPHN Metadata Catalog, a system designed to enhance the FAIRness of biomedical datasets within the SPHN. By providing structured, machine-readable metadata and interactive visualization, the catalog addresses key challenges in dataset discoverability, interoperability, and reusability while maintaining data privacy.

Data repositories like the European Genome-Phenome Archive (EGA) [58] or general-purpose open research repositories like Zenodo [59] can provide insights into the hosted content directly. In contrast, such direct provision of data is not an option for clinical or health-related data due to its potentially sensitive nature and the need to protect patient privacy. The SPHN Metadata Catalog addresses this challenge by providing researchers with rich descriptive metadata as well as information on dataset structures, data types, coding systems, and aggregated counts without revealing sensitive data. This allows insights into the data and supports secondary analyses. It provides its users with a solid information basis to decide whether datasets identified are suitable for the intended purpose. This stimulates new collaborations by making dataset characteristics transparent and accessible without exposing critical information. Users can reach out to the contact specified in the metadata to request access to data identified as fit for purpose (Figure 1).

Importantly, the catalog reflects the multilevel nature of data provenance, which is critical for auditing, verification, and reuse. Provenance is captured at the dataset level primarily through metadata describing origin, stewardship, and data sources and other characteristics, including data provider, responsible organization, source information (eg, Clinical Data Platforms, Biobanks, or OMICS), creation and update dates, versioning, and links to related resources. This enables users to quickly understand the context and scope of the dataset and, at the same time, assess its origin and stewardship even before requesting access.

For datasets that follow the SPHN RDF Schema, additional provenance can also be represented at the level of individual data elements via attributes such as source system and semantic mapping [60]. Because not all datasets in the catalog conform to the SPHN RDF Schema, the granularity of provenance information may vary across resources. In such cases, provenance is primarily represented as dataset-level metadata. This flexible approach supports transparency while accommodating dataset heterogeneity.

Some adoption barriers remain, as not all researchers have sufficient resources, awareness, or incentives to share their metadata with third parties. These aspects represent general issues of metadata provision, not limitations specific to the SPHN Metadata Catalog. In the context of SPHN, projects must deliver metadata at least at the closure of the project as part of the funding regulations.

For the quantitative metadata, the SPHN Metadata Generator tremendously reduces manual effort by automating the generation of count information while ensuring consistent quality and updates. However, the set of qualitative metadata needs to be collected through manually filled templates. Combined with the need for subsequent curation and quality control by the SPHN team, this process is resource-intensive. Future work will focus on streamlining qualitative metadata collection through structured and standardized mechanisms like restriction of fields to specific value sets and automated validation, thereby supporting both the metadata providers and curation teams.

The schema of the SPHN Metadata Catalog fosters metadata interoperability by building on common metadata standards like DCMI Metadata Terms, DCAT, and FDP-O while also accommodating metadata elements that are of significance to SPHN. To further strengthen interoperability and alignment with other cataloging initiatives, metadata elements are harmonized with, and can be mapped to, additional profiles and schemas such as DCAT AP, DCAT-AP CH, and the Health-RI Metadata Schema. The SPHN Metadata Catalog Schema is expected to evolve alongside changing research and clinical requirements; therefore, each schema release is versioned, documented, and maintained to ensure traceability, while backward compatibility is supported through defined transformation rules that allow previously generated metadata to remain interoperable with updated versions. This approach also accommodates the ongoing evolution of frameworks such as HealthDCAT-AP, ensuring continued alignment with emerging standards. This alignment enables the exchange and reuse of metadata across different data catalogs, research infrastructures, and data platforms while maintaining consistent semantics.

Harvesting metadata from the SPHN Metadata Catalog into existing repositories, such as the national I14Y Interoperability Platform of the Swiss Federal Office of Statistics [24], has been established. Additionally, interoperability with the currently established European Health Data Portal [61] will enable federated queries across different catalogs. Both initiatives foster broader data integration and reuse.

The SPHN Metadata Catalog consists of 2 core components that jointly strengthen the discoverability and reuse of SPHN datasets: the visual exploration tool SPHN Schema Scope and the standardized FDP enabling interoperable metadata access. In addition, a SPARQL endpoint supports advanced querying capabilities. Together, these components provide both intuitive and programmatic access to structured metadata, allowing users to explore, retrieve, and interrogate information in a flexible and consistent way.

The underlying technologies support interoperability not only within SPHN but also at the national and international levels, promoting responsible data reuse and enhancing the visibility and impact of SPHN projects and beyond.