Meetings
BioHackSWAT4HCLS 2026
BioHackathon Europe 2026
BioHackathon Germany 2026
DBCLS BioHackathon 2026
ELIXIR INTOXICOM
Recent preprints
-
DBCLS BioHackathon 2026 report: Extending TogoMCP beyond RDF Portal while making its schema guides check their own answers
TogoMCP is a Model Context Protocol (MCP) server that lets large language model (LLM) agents query life-science knowledge graphs in SPARQL, guided by per-database schema documents called MIE(Metadata Interoperability Exchange) files. During the DBCLS BioHackathon 2026 (BH26), the TogoMCP group set out to extend the server, refine the MIE files, add databases, learn what makes a good SPARQL example, and turn use cases into reusable agent skills. In the first six days of the event we published eleven releases (v2.12.2 to v2.20.0) and grew the catalogue from 37 to 45 databases: Fanta.bio, WikiPathways, IDSM, PubCase Finder, LIPID MAPS, SwissLipids and MarpolBase, plus BH26Microbes, an experimental dataset of KEGG Orthology assignments for 57.6 million prokaryotic proteins built for the BioHackathon. Five of them are served from endpoints outside RDF Portal,and a sixth from a QLever engine inside it rather than Virtuoso, and they broke assumptions the rest of the corpus had taught, such as named-graph pinning, federation with SERVICE, literal typing, and even that a result set is complete. For the first time, MIE files were written by people other than the server’s maintainer, one of them by the maintainer of the database it describes, and a second group member built TogoCX, a companion MCP server that returns database edges with instructions on how their claims may be stated. We made every worked example in the MIE corpus (now 420) assert its recorded result against the live endpoint in continuous integration.The first full comparison found drift that execution-only checks had missed, most seriously a NANDO release that made an example join silently miss about 88% of mapped diseases. Onboarding the new databases uncovered a series of quantified “silent wrong answers”, queries that return plausible results rather than errors, including a category count inflated 3.1-fold by an upstream data defect and a gene enumeration that returns 10,000 of 18,080 rows without an error or awarning. We also added two PubCaseFinder tools for phenotype-driven rare-disease diagnosis support, began serving analysis workflows (agent skills) from the server, fixed failure modes found in production call logs, made TogoID errors suggest working conversion routes, and prepared a LOTUS natural-products graph for hosting on RDF Portal. We argue that for LLM-facing schema documentation, “the example still runs” is not evidence that it is still correct. -
DBCLS BioHackathon 2026 report: Open, Reproducible Extraction of Data-Entity References from the Biomedical Literature – with Entry-Level Resolution via Identifiers.org / TogoID
Funders and data producers need to know not just which databases are cited in the literature, but whether a cited accession actually exists and how it was used. Making that reuse visible is itself an incentive for the wider research community to deposit data, by giving concrete, attributable evidence that public deposits are genuinely reused and contribute to science. We report on a pipeline, built during the DBCLS BioHackathon 2026, that extracts database identifiers from PMC full text, verifies each one against source-of-truth APIs (TogoID/RDF Portal, NCBI E-utilities, EBI Search), and classifies confirmed accessions into one of five credit/provenance classes. On a working slice of 10,896 PMC documents, the pipeline confirms 19,786 (paper, accession) links across 1,352 papers and 18,159 distinct accessions, though a single namespace (insdc) accounts for roughly 70% of that volume. An independent comparison against Europe PMC’s own accession annotations over the same slice shows 74.6% agreement at the individual-citation level, with a bimodal split at the per-document level: 64% of documents that either system found anything in agree at 90-100%, while 16% agree at only 0-10%, a pattern that needs manual citation and analysis. The pipeline, its Europe PMC comparison, and a static, backend-free dashboard reporting the results are complete and reproducible for this slice. A reference-role (generated/used/mentioned) classifier and a 221-item human gold standard evaluating it were also completed, reaching 0.80 macro-F1 with a near-zero generated-vs-used confusion; extending coverage and running against the full corpus are left as future work. -
Logica, an intelligent platform for semantic interoperability
Clinical AI depends on both the relevance of the information it receives and the meaning of that information. Logica is an interoperability platform that separates clinical meaning from physical data representation and separates language-model interpretation and reasoning from deterministic retrieval. A first language model interprets a clinical question through the COOL semantic graph. Explicit information requirements are compiled into Clinical Quality Language (CQL) over COOL, a logical representation of a query, bound to a physical backend (e.g., as a FHIR query or SQL query), and executed in code. Returned records are normalized into COOL evidence with source provenance before a second language model reasons over them. Deterministic calculations and citation-identifier validation make the resulting answer inspectable. We describe this architecture through a reported creatinine demonstration for a synthetic patient (Demo Patient A), in which a record containing 204 FHIR resources yielded 69 evidence items. The demonstration illustrates the pipeline and its retained artifacts as well as how the system works; however, further evaluation is needed to determine whether it actually improves the saliency of relevant clinical information or clinical reasoning accuracy compared with other approaches. The research hypothesis is that a sufficiently constrained semantic model improves context relevance, reasoning accuracy, and verifiability. Separating logical expressions from backend bindings also provides a basis for sharing clinical logic across institutions. -
DBCLS BioHackathon 2026 report: QPX: Pathway analysis environment for non-model organisms
We report progress on Quest for Pathways with eXpression (QPX), a pathway analysis environment for non-model organisms, developed during the DBCLS BioHackathon 2026 (BH26). Building on prior BH23/BH25 work, we demonstrated PathLift, a tool that lifts WikiPathways GPML pathways to other species via ortholog mapping, using an Arabidopsis-to-rice mitochondrial complex I example. We also created annotated pathway maps in QPX integrating transcriptome/proteome data for hypoxic stress response, and transcriptome/metabolome data for the non-model organism Symplocarpus renifolius, which lacks an established gene ID system. This clarified data-formatting requirements for linking QPX maps to expression tables and surfaced a string-identifier handling issue in QPX’s backend, reported upstream. Future work includes reproducible ChEBI ID assignment, automated pathway digitization, and compound-centered functional networks from PubChem/PubTator3 data. -
DBCLS BioHackathon 2026 report: Variant representation in RDF for precision medicine
For the life science semantic web community, the provision of genomic variants ready for tools and applications is essential to deliver on the promise of precision medicine. As part of the DBCLS BioHackathon 2026, we here report our efforts on how best to provide variant data in RDF for downstream patient-clinical trial matching applications. -
Building InterMine databases from RDF Portal
InterMine databases such as HumanMine integrate many biological data sources, but each source needs its own loader, and keeping those loaders current is costly. The public HumanMine has not had a data release since February 2022. RDF Portal, operated by the Database Division for Life Science, now serves many of the same sources as reviewed RDF, and describes each dataset with an rdf-config model. At the DBCLS BioHackathon 2026 we developed rdfc2im, a tool that maps rdf-config models onto the InterMine data model, keeps the mapping as reviewable data with the evidence for every row, uses RDF Config to generate SPARQL queries to acquire data from RDF Portal, and writes the data into a form that InterMine’s standard loader integrates without new code. We used it to build a working demonstration HumanMine from nine sources and a panel of 113 food- and drug-metabolism genes, with NCBI Gene, the Gene Ontology and Reactome loaded in full. The mine supports search, template queries and list analysis. Real loads found problems that static checks did not, most of them related to the way objects from different sources are identified and merged, and to the limits of public SPARQL endpoints. This report describes our proof of concept. We list what remains to be done before rdfc2im can rebuild a full HumanMine from each RDF Portal release. -
OSIRIS-MCP: Towards secure conversational access to research information in OSIRIS
Research information systems contain structured information about researchers, organisational units, projects, publications, and other research activities. However, answering ad hoc questions for reporting, management, or public communication often still requires manual searches, exports, and synthesis. We developed a prototype Model Context Protocol (MCP) connector for OSIRIS, an open-source research information system, during the DBCLS BioHackathon 2026. The connector gives language-model clients read-only access to a deliberately restricted set of typed search and retrieval tools. It combines a compact, evidence-centred data representation with instance-specific metadata discovery, bounded pagination, dedicated API-client permissions, safe error handling, and either OAuth or static API-key authentication. The work also introduced a general API-client registry in OSIRIS, allowing integrations to receive independently revocable credentials and explicitly assigned permissions. A containerised deployment was validated against a local OSIRIS instance containing synthetic data and tested interactively with MCP clients. The prototype demonstrates that conversational access to institutional research information can be implemented without exposing complete database records or making the language model responsible for access control. Evaluation with real institutional data and deployment guidance remain necessary before production use.
- •
- 1
- 2