This deliverable describes the EPND Federated Discovery Pilot, developed by Work Package 1 to test dataset discovery across multiple catalogues. The pilot established an operational network spanning five nodes and 78 cohort or dataset entries, demonstrating how a single query portal can identify research resources held across participating systems.

The pilot combines core dataset metadata with pre-aggregated record-level information. Examples include participant age characteristics, counts by sex and follow-up duration. This approach was selected to provide more informative searches than basic catalogue descriptions while avoiding direct queries against individual participant records. The pilot focused on datasets; biosample collections were excluded because identifying suitable samples generally required more detailed discussions with biobank operators.

The technical implementation uses the GA4GH Beacon-2 API framework, a shared metadata and aggregate-data specification, and JSON representations. The Beacon-2 interface was adapted to support filters relevant to the pilot, including participant numbers, age ranges, country, dataset type and study dates. A common specification defines the information and formats required for queries across the network.

Most nodes used Café Variome v3 software, adapted to support the EPND models and discovery interfaces. BMD Software implemented a separate Beacon-2 service to make EMIF-AD catalogue entries discoverable. The central portal combines results from these different implementations and uses the EPND Hub’s Keycloak service for user authentication.

The report also describes operational limitations and options for further development. Where a node cannot apply a requested filter, the interface warns users that the corresponding condition was omitted. Integration with the main EPND Catalogue remained a subsequent design decision, with options for combined or separate discovery interfaces.

The pilot provides a tested technical foundation for expanding federated discovery. It supports identification of relevant datasets; access to the underlying data remains a separate process.

For more information, find the full Deliverable report here.