# PDR 009: Publish typed biomedical knowledge pages

- HTML version: https://robbiepalmer.me/projects/personal-knowledge-graph/pdrs/009-typed-biomedical-knowledge
- Project: Personal Knowledge Graph
- Status: Accepted
- Opened: 2026-11-11
- Decided: 2026-11-11

## Context

Projects such as Pathologize use terms whose meaning cannot be recovered from
the project name or a dataset summary. A reader may need to understand
pathology, genomics, FFPE, H\&E staining, immunohistochemistry, fluorescence,
Gleason grading, MSI, or PD-L1 before the work makes sense.

These terms do not all describe the same kind of thing. Pathology and genomics
are scientific fields. FFPE is a specimen preparation method. H\&E staining is
an assay method. Genomic data is a data type. MSI and PD-L1 can be biomarkers.
Gleason grading is a grading system. Multiple-instance learning is a data
science technique, even when a pathology project uses it.

Treating every term as a pathology topic would make genomics and other
molecular evidence subordinate to one field. Treating every term as an idea
would hide distinctions that readers need and that projects should state
plainly.

## Decision

Publish biomedical knowledge as typed pages in the existing connected graph.
The first vocabulary will distinguish:

| Class                       | Examples                           |
| --------------------------- | ---------------------------------- |
| Biomedical field            | Pathology, genomics, proteomics    |
| Specimen preparation method | FFPE, fresh frozen                 |
| Assay method                | H\&E staining, IHC, sequencing     |
| Biomedical data type        | Histology images, genomic data     |
| Biomarker                   | MSI, PD-L1                         |
| Grading system              | Gleason grading, ISUP grade groups |

Image and spatial concepts may gain concrete classes when published content
needs them. The graph will not add a generic biomedical subject or reference
page as a fallback.

Biomedical fields explain the disciplines themselves. Pathology,
histopathology, molecular pathology, computational pathology, genomics,
transcriptomics, proteomics, and bioinformatics can therefore have their own
pages. A field page must not absorb the evidence or techniques that the field
uses. Genomics and genomic data remain separate, just as pathology and a
pathology assay remain separate.

General analysis techniques stay in the existing idea vocabulary.
Multiple-instance learning and spatial analysis are ideas that biomedical
projects may use. Multi-omics is also an idea when it describes an integration
approach. Their pages should explain the technique without assuming a medical
use case. Biomedical pages and projects can link to them for context.

Each biomedical entity has one stable local identity and one canonical page
under `/biomedicine/<slug>`, with a plain Markdown twin at the same URL plus
`.md`. The page states its class. Changing a title or adding an external
ontology mapping does not change its local identity.

Every page records the sources used for its claims and the date on which the
content was last reviewed. External ontology identifiers may map the local
entity to established vocabularies, but they do not replace the site's
editorial identity or prose.

Projects may link to the fields, preparations, assays, data types, biomarkers,
grading systems, and general ideas they use. Blog posts may explain or discuss
the same canonical entities. Dataset and data-product pages may link to them
later without becoming the source of their definitions.

Medical images and whole-slide viewers should appear where they help a reader
understand visual evidence. Each example must use a real, licensed source and
show its attribution. A viewer also needs a useful non-interactive fallback.
Generated imagery must not stand in for clinical or research evidence.

## Consequences

Readers can learn the domain before following a project into datasets,
benchmarks, or implementation details. The same explanation can support more
than one project and more than one kind of evidence.

Authors must decide what a term is before publishing it. That adds work, but it
prevents field names, laboratory methods, measurements, and analysis
techniques from collapsing into one list of tags.

Biomedical claims can age or vary by authority. Pages need named sources and a
visible review date. Interactive evidence also brings licensing, attribution,
accessibility, storage, and maintenance work.

The initial vocabulary will not cover every biomedical entity. A new concrete
class requires a small ontology decision. It must not enter through a generic
subject type.

## Alternatives considered

### Make pathology the top-level domain

This would fit the first Pathologize pages. It was rejected because genomics,
transcriptomics, proteomics, and molecular biomarkers are not subtypes of
pathology. Pathology remains a field within the broader biomedical section.

### Publish every term as an idea

This would reuse an existing route and schema. It was rejected because a
scientific field, specimen preparation, assay, data type, biomarker, and
grading system have different meanings and relationships. Ideas remain the
right home for domain-independent analysis techniques.

### Add one biomedical subject type

This would shorten the schema and loader work. It was rejected because the
type would only say where a page appears on the site. It would not say what
the entity is or constrain the relationships that can point to it.

### Let datasets define the terminology

This would keep explanations close to their first use. It was rejected because
the same concept recurs across datasets, projects, and posts. Dataset metadata
should link to a canonical definition rather than duplicate one.

## Related records

- Project: [Personal Knowledge Graph](https://robbiepalmer.me/projects/personal-knowledge-graph.md)
- Informed by ADR: [ADR 033: In-Memory Content Graph](https://robbiepalmer.me/projects/personal-knowledge-graph/adrs/033-content-graph-indexes.md)
- Informed by ADR: [ADR 038: Content Graph Explorer Rendering](https://robbiepalmer.me/projects/personal-knowledge-graph/adrs/038-content-graph.md)
- Informed by ADR: [ADR 060: Model financial and business subjects as typed content graph nodes](https://robbiepalmer.me/projects/personal-knowledge-graph/adrs/060-typed-financial-and-business-subjects.md)
- Implemented by ADR: [ADR 061: Model biomedical knowledge as typed content graph nodes](https://robbiepalmer.me/projects/personal-knowledge-graph/adrs/061-typed-biomedical-content-nodes.md)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt