BioAgents Metadata Harvesting
BioAgents reads deposited papers and datasets and proposes structured metadata: bibliographic fields, entities such as genes, species, and methods, and links into the Ekayana knowledge graph. The data owner reviews and approves every record before it is published.
Structured metadata is expensive to produce by hand
A research paper is a terrible database. The knowledge it holds, this gene associates with that disease, this method was applied at that site, sits in prose, findable only by someone who already knows to read that particular paper. Structured metadata is what makes discovery and reuse possible, and manual curation rarely keeps pace with what an institution deposits.
Curating a rich record by hand takes expert time, and that time competes with everything else a repository team owes its depositors. As deposit volume grows, records get thinner. Thin records are hard to find.
Everything downstream inherits that quality. Search, citation tracking, and cross-institution analysis can only work with what the records actually say. A persistent did:bio identifier carrying three fields of metadata resolves perfectly well and still goes unread, which applies equally to datasets that keep their DOIs in Dataverse or Zenodo and gain a did:bio alongside them.
A three-stage extraction pipeline
BioAgents is the AI layer of the Ekayana platform. You hand it a research paper or dataset; it returns structured metadata, recognized entities, and RDF triples ready for the knowledge graph. The point is not to summarize documents. It is to make a corpus of them queryable as data, and each stage below narrows the representation a little further.
Document processing
GROBID
Text and structure are pulled out of the PDF: sections labelled, citations segmented, tables and references normalized. Scientific PDFs are hostile territory, and a clean segmentation is most of the work.
Knowledge extraction
Entities, relations, claims
Entity recognition runs over the cleaned text for genes, proteins, species, diseases, compounds, methods, and locations. Relations between them are detected, claims are extracted together with the passage that supports them, and the bibliographic fields are read off the document.
Knowledge representation
RDF and JSON-LD
Extracted facts become RDF triples and JSON-LD documents that link into the existing knowledge graph. From that point the contents of the paper can be queried alongside every other record.
Pipeline endpoints
Processing runs on task workers in the Eliza OS agent framework: a submission returns a task ID immediately, and a forty-page PDF never blocks a connection. Four REST endpoints cover the flow, and the Ekayana Console uses the same ones.
| Endpoint | Method | Purpose |
|---|---|---|
/api/bioagents/process | POST | Submit a paper or dataset for extraction |
/api/bioagents/status/{task_id} | GET | Check the status of a processing task |
/api/bioagents/metadata | POST | Retrieve extracted metadata for owner review |
/api/bioagents/query | POST | Query the knowledge graph in plain language |
BioAgents endpoints sit behind the same authentication and UCAN authorization as the rest of the API, and connect to the Bio-DID-Sequencer core so that approved metadata can be written into DID documents.
Submit, poll, review, publish
Four steps take a deposited document to a published metadata record. The Ekayana Console guides the whole flow, or you can drive the pipeline from your own systems over plain HTTP; the documentation covers every endpoint with request and response examples.
Submit
Upload the paper or dataset to IPFS and submit its CID from the Ekayana Console, or call
POST /api/bioagents/processdirectly. A UCAN capability check runs before the pipeline sees the file.Poll
Processing is asynchronous. Poll
/api/bioagents/status/{task_id}until the task completes, or watch progress in the Console.Review
Fetch the extracted metadata and review it as the data owner: correct what is wrong, drop what does not belong, approve what is right. Nothing has been published at this point.
Publish
The approved record is written into the did:bio DID document, and the entities it names link into the knowledge graph.
Submit, poll, retrieve (JavaScript)
// 1. Send an uploaded paper through BioAgents
const res = await fetch('https://api.ekayana.com/api/bioagents/process', {
method: 'POST',
headers: {
Authorization: `Bearer ${token}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
file_cid: 'QmXg9Pp2ytZ14xgK35M6iTC2Vz6jR9zYgooNp2UHPTMnPN',
title: 'CRISPR-Cas9 Gene Editing for Neurodegenerative Diseases',
authors: ['Jane Smith', 'John Doe'],
doi: '10.1234/example.2026.001',
}),
});
// 2. Processing is asynchronous. Poll the task until it completes.
const { task_id } = await res.json();
const status = await fetch(
`https://api.ekayana.com/api/bioagents/status/${task_id}`,
{ headers: { Authorization: `Bearer ${token}` } },
);
// 3. When it completes, fetch the extracted metadata for the owner to review
const metadata = await fetch('https://api.ekayana.com/api/bioagents/metadata', {
method: 'POST',
headers: {
Authorization: `Bearer ${token}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({ task_id }),
});Nothing is published without owner review
BioAgents publishes nothing on its own. Extracted metadata is staged for the data owner, who reviews, corrects, and approves it before any of it reaches the DID document or the knowledge graph. Access is capability based, so a paper you have not shared cannot be processed by someone else's agent: the UCAN check happens before the pipeline sees the file. Data is encrypted in transit and at rest, access is audit logged, and GDPR rights apply throughout.
What BioAgents extracts
The extraction suite is included with every Ekayana membership, and individual researchers get basic BioAgents processing on the free Researcher tier. The fee schedule sets out what each tier covers.
Descriptive record
- Bibliographic metadata
- Authors, affiliations, publication details, and citation lists, taken from the document itself.
- Methods and instruments
- Experimental methods, protocols, and instrumentation named in the text. These are the fields that decide whether a dataset is reusable in another study.
- Locations and context
- Geographic locations, sampling sites, and temporal context, which matter as much for heritage documents as for field studies.
Entities and links
- Biological entities
- Genes, proteins, species, diseases, and compounds, recognized in context and normalized so the same entity resolves identically across a corpus.
- Relationships and claims
- Which gene associates with which disease, which compound inhibits which pathway, each claim carried alongside the evidence cited for it.
- Knowledge-graph triples
- Every approved fact lands as an RDF triple linked to its did:bio, so a set of deposits becomes one queryable graph.
Where harvesting fits
BioAgents is one layer of the platform. Identifiers are registered beneath it, discovery runs on what it extracts, and repository integrations decide when it runs.
Register DIDs
Every harvested record attaches to a persistent did:bio anchored on the Solana registry.
Enable Discovery
The knowledge graph BioAgents populates is what makes semantic search and plain-language queries possible.
Integrate Workflows
Repositories can run BioAgents at deposit time, while the metadata context is still fresh.
Membership includes the extraction suite
Institutional Members run BioAgents under their own account in the Ekayana Console and get knowledge-graph views over the records it produces. Individual researchers can start on the free Researcher tier without an institutional agreement.