Gene Sets

dismech consumes curated gene sets and their GO interpretations from monarch-initiative/genesets (MSigDB/KEGG/WikiPathways/Hallmark pathways and cell-type signatures). Two things work with them:

1. Cite a gene set as evidence — each set is a structured reference source (MYGENESET:<id>) cached as line-oriented markdown.
2. Align a gene set's curated biological processes to a disease pathograph — just genesets-align scores how well a disorder's curated mechanism covers the gene set's GO terms.

just targets

Structured source: MYGENESET:

src/dismech/structured_sources/mygeneset.py ingests 100 gene sets into references_cache/MYGENESET_<id>.md. Identity is keyed on mygeneset.info: the reference id is MYGENESET:<_id> (the bare set name, e.g. KEGG_ASTHMA). The pin lives in data/genesets/MANIFEST.yaml (genesets repo commit + mygeneset/mygene base URLs); data/genesets/interpretations/ and membership/ are fetched, not committed.

Each cache file has these sections, with every row a quotable evidence snippet::

• Description, Context (disease/tissue MONDO/UBERON),
• Curated GO interpretation — GO id | label | aspect | role | confidence,
• Members — symbol | hgnc:id (NCBIGene→HGNC resolved via mygene.info; NCBIGene fallback only for loci with no HGNC),
• Source — genesets commit + mygeneset + upstream gene_set_id xref.

Citing a row as evidence (a gene-set row is a lead, not mechanism — back disease claims with a primary PMID too):

evidence:
- reference: MYGENESET:KEGG_ASTHMA
  supports: SUPPORT
  snippet: "GO:0045064 | T-helper 2 cell differentiation | biological_process | core_process | high"
  explanation: The curated KEGG_ASTHMA interpretation lists Th2 differentiation as a core process.

The gene_sets slot

A Disease declares its gene sets via the gene_sets slot (GeneSetAssociation: gene_set + relationship + note). This is the precise, curated disease↔set link — membership and the GO interpretation stay in the cache file, so the slot never re-duplicates genes.

gene_sets:
- gene_set: MYGENESET:KEGG_ASTHMA          # → references_cache/MYGENESET_KEGG_ASTHMA.md
  relationship: CANONICAL_PATHWAY
  note: KEGG legacy asthma pathway; overlaps the Th2 + inflammatory arms.

GeneSetRelationshipEnum: CANONICAL_PATHWAY, CELL_TYPE_SIGNATURE, PERTURBATION_SIGNATURE, DISEASE_SIGNATURE, OTHER.

30 disorders declare gene_sets (see the disease list above). Each declared link is a precise, per-entry mapping — e.g. KEGG_COLORECTAL_CANCER is on Colon_Adenocarcinoma, while metastasis/EMT-specific sets belong on the metastatic entries.

BP alignment

src/dismech/genesets_align.py compares a gene set's curated GO terms to a disorder's biological_processes + cellular_components + protein_complexes, matched over GO via OAK:

• Match status: EXACT and DESCENDANT (pathograph term is more specific) count as represented; ANCESTOR (pathograph has only a more general term) is shown but does not count.
• Role-weighted: corroboration is scored over the set's core_process/ core_component BPs; nonspecific/false_association BPs are expected absent and only lint if they appear in the pathograph.
• Output: per-BP status, a corroboration fraction (core BPs represented), and the core-BP gap list.

genesets-align-all runs this across the catalog, preferring the explicit gene_sets links (marked declared) and falling back to a one-disorder-per-MONDO guess (marked mondo?) for sets no disorder declares. Recurring gaps it surfaces: MHC class II protein complex (a GO cellular-component, modeled via CL in dismech) across Asthma/RA/T1D; mitochondrion + neuron apoptotic process across neurodegeneration; regulation of cell population proliferation across the cancer pathways.

Coverage

100 ingested sets; 56 are disease-context. 30 are wired to a disorder via gene_sets. The rest either have no dismech disorder yet, or are cell-type / process / signature sets that attach to phenotypes or modules rather than a single disease. The IL6→JAK/STAT MEDICUS set is a signaling pathway, not a disease — it belongs on a mechanism module.