Cytognosis maintains a single LinkML schema that unifies bibliographic metadata, scholarly graph data, biomedical entities, research artifacts (code, data, models, workflows, protocols), licenses, and personal annotations — and is interoperable with SemOpenAlex SPARQL, the UMLS Metathesaurus, the NCATS Translator ecosystem, and RO-Crate packaging.
The canonical schema YAML lives in the Infrastructure and Tooling workspace folder rather than this repo, because it is regenerated routinely by the codegen pipeline and downstream artifacts (Pydantic, JSON Schema, OWL, SHACL, SQL DDL) are large. This document is the authoritative description of what the schema is for, where to find it, and how it is structured.
| Artifact | Location |
| --- | --- |
| Current schema (v0.4.0) | Infrastructure and Tooling/cytognosis_scholarly_kg_v0.4.0.yaml (workspace) |
| Previous schema (v0.3.0) | Infrastructure and Tooling/cytognosis_scholarly_kg.yaml (workspace) — retained for reference |
| Codegen outputs | regenerated on demand into Infrastructure and Tooling/cytognosis_scholarly_kg/ (Pydantic, JSON Schema, SHACL, OWL, SQL DDL) |
When the schema is promoted to a public release we will publish it under cytognosis/scholarly-kg-schema with a SPDX license, FAIR identifier (DOI via Zenodo), and a Hugging Face data-card mirror.
Several existing schemas overlap, none cover everything we need, and stitching them in code is brittle:
A LinkML schema lets us import all of the above as YAML modules and add typed Cytognosis-specific extensions on top, while emitting Pydantic / JSON Schema / OWL / SHACL / SQL DDL automatically.
The 2026-04-11 v0.4.0 release added typed artifact relationship edges, RO-Crate alignment, computational workflows, formal parameters, instruments, protocols, presentations, and several enums.
Three enums encode the relations we care about between papers and the artifacts they cite, produce, or extend.
``text
PaperCodeRelationType
implements (schema:isBasedOn)
supplements (schema:hasPart)
depends_on (prov:used)
extends (pav:previousVersion)
documents (codemeta:referencePublication)
reproduces (cito:replicates)
PaperModelRelationType
proposes (prov:wasGeneratedBy)
uses (prov:used)
fine_tunes (prov:wasDerivedFrom)
evaluates (cito:citesAsAuthority)
distills (prov:wasDerivedFrom)
extends_architecture (pav:previousVersion)
PaperDatasetRelationType
creates (prov:wasGeneratedBy)
uses (cito:usesDataFrom)
uses_for_training
uses_for_testing (cito:citesAsDataSource)
uses_for_validation
uses_as_benchmark (cito:citesAsAuthority)
augments (prov:wasDerivedFrom)
curates (prov:wasDerivedFrom)
`
Each relation is captured as a concrete subclass of an abstract ArtifactRelationship:
Every concrete relationship carries source / target references, the typed predicate, supporting evidence_text, and a confidence score.
(schema:Dataset) — the RO-Crate root entity, with typed has_part slots for papers, code, datasets, models, workflows, protocols, presentations, and instruments. mixin — conforms_to, sd_publisher, sd_date_published, credit_text, cite_as. enum — base, workflow, workflow_run, process_run, provenance_run, workflow_testing, five_safes, arc. (bioschemas:ComputationalWorkflow) — workflow language, inputs / outputs as FormalParameter. (bioschemas:FormalParameter) — direction, type, default value. (schema:Thing) — manufacturer, model, serial number. (schema:CreativeWork) — protocol type, steps, reagents. (schema:PresentationDigitalDocument) — format, event name, slide count., WorkflowLanguage (cwl, wdl, nextflow, snakemake, galaxy, …), ParameterDirection, ProtocolType, ResearchObjectProfile.
New CURIE prefixes
| Prefix | Namespace |
| --- | --- |
|
pav: | Provenance, Authoring, Versioning ontology |
| bioschemas: | bioschemas.org |
| ro: | Research Object Ontology |
Schema statistics (v0.4.0)
38 enums
~95 classes (including 13 mixins)
250+ slots
How the schema is used
| Surface | How it consumes the schema |
| --- | --- |
| Zotero / Drive integration | The
Extra field on Zotero items uses keys aligned with the schema slots (drive-pdf:, code-original:, etc.). See paper-library-architecture.md. |
| Monday.com Resources workspace | Each entity board's columns map 1:1 to schema slots. Relationship boards instantiate the typed enums above. See monday-resource-boards.md. |
| Future Neo4j knowledge graph | Nodes and edges in the objective layer are produced by codegen against this schema. The personalization layer is W3C Web Annotation nodes layered on top. |
| NCATS Translator Knowledge Provider | The schema's Biolink-aligned classes let us register Cytognosis as a KP exposing typed paper / code / model / dataset / workflow / protocol relationships. |
| RO-Crate packaging | ResearchObject + HasROCrateMetadata mixin produce conformant RO-Crate metadata for any submitted release. |
| SSSOM cross-ontology mapping | The schema imports sssom_schema; an EntityMapping subclass of Mapping links to ResearchObject for provenance. See sssom-cross-ontology-mapping.md. |
Open follow-ups
Promote the schema to its own public repo with a Zenodo DOI once v0.5.0 settles.
Refactor v0.5.0 against the hybrid-layout pattern in chapter 13 of the LinkML+KG playbook.
Wire linkml validate and sssom validate into CI for any PR that touches the schema or the SSSOM TSVs.
Cross-references
paper-library-architecture.md — Layer 0–3 sovereign library that is the operational front end of this graph.
sssom-cross-ontology-mapping.md — how biological entities are harmonized across vocabularies before ingest.
monday-resource-boards.md — Resources workspace that mirrors the schema's entity types.
linkml-kg-playbook.md` — hands-on playbook covering LinkML, Biolink, BioCypher, Koza/Monarch, OLS4 SSSOM, CELLxGENE harmonization.