Senescent cells in DisMech: paper-grounded representation pilot

Date: 2026-09-20. Repository baseline: 1cc412fddde5aacb0cc66311a96e25d67465fdb9. Status: local exploration; no production schema/KB changes, commit, push, PR, or publication.

Recommendation: use the existing cell descriptors, mechanism nodes, anatomical descriptors, and observational readouts first. Preserve the identity of a fibroblast, AT2 cell, or T cell separately from the claim that a particular population is senescent. Represent induction, maintenance, secretion, escape from arrest, immune evasion, clearance, and accumulated burden as different claims when the evidence distinguishes them. A small optional, cell-scoped state assertion could improve queries across mechanisms, but this paper does not establish an ontology definition or an operational marker rule for such a field. Do not introduce it into production on the strength of this pilot alone.

The main obstacles are semantic: PATO's senescent is a broad aging quality, GO's cellular-senescence definition specifies irreversible arrest, and the review explicitly allows context-dependent escape. A syntactically valid annotation can still overstate the biology. The examples and validation deliberately expose that boundary.

Source identity and reading record

[P] Elina Shakur, Abraham Jacobs, Rituparna Ghosh, and Matthew J. Yousefzadeh. Escape and evasion: when immunosurveillance of senescent cells goes wrong. Frontiers in Genetics 17:1882818, published 7 July 2026. DOI: 10.3389/fgene.2026.1882818. Article type: Mini Review, not original experimental research.

Title, all four authors, article type, journal, volume/article number, DOI and publication date were read from page 1 of the uploaded PDF, not inferred from its filename. Page 1 also records received 16 May, revised 12 June, and accepted 18 June 2026. PubMed independently confirms the identity as PMID:42518598, with PMCID PMC13384439. PubMed/cache title punctuation adds a terminal period; this is not a different paper.

Uploaded source: /Users/cjm/.hermes/cache/documents/doc_ac0569aee4af_fgene-17-1882818.pdf. Nine pages, printed pages 01–09; SHA-256 4eec4659227597d37717344de0e6fe527861f78f33cb9834e4f279dd38a53701. The PDF was read through page 9, including references and declarations. Figures 1 and 2 were inspected visually, not inferred from extracted text alone. The PDF was left unchanged. Full-text extraction was used for reading, not copied into this bundle.

What the paper supports, and what is a design inference

Each row states a passage-level basis. The modeling consequence is this pilot's inference, not a schema recommendation made by the authors.

The paper supplies no DisMech schema, CL/PATO mapping, universal marker panel, threshold for declaring senescence, or dataset of individually annotated cells. The identity/state separation proposed here is a modeling choice motivated by these examples, not an experimental conclusion about immutable lineage identity. Do not infer that a senescent cell preserves all lineage properties indefinitely.

Source defects that matter to curation

1. p. 5 says “HLE-A” in the fibroblast/endothelial-cell sentence, while citing Pereira et al. (2019). The cited primary paper is Senescent cells evade immune clearance via HLA-E-mediated NK and CD8+ T cell inhibition. Its abstract identifies HLA-E on senescent dermal fibroblasts and the inhibitory receptor NKG2A; it describes human skin observations and in-vitro blockade. Use HLA-E for a claim verified against Pereira, and retain the original spelling if quoting Shakur. Neither HLA-A nor an invented “HLE-A” entity is justified.
2. Figure 2, p. 4 labels senolytics “Induces anti apoptotic pathways,” while its caption describes inducing apoptosis. This is an internal figure/caption conflict. The existing module's independently cited senolytic pattern is selective killing through disruption of survival networks. Do not encode the figure label as a causal mechanism. No graphic or source text was corrected in place.
3. The review's high-level grouping of immune cells and schematic receptor/ligand drawing are not ontology axioms. In particular, do not infer cellular lineage, receptor localization, or antigen-presentation restrictions solely from the drawing. Primary-study methods would be required for detailed interaction curation.

The Demaria acute-wound example was checked against PMID:25499914: the underlying experiments used mice. The DOI fetch produced metadata without content after a publisher PDF 403; fetching the verified PMID succeeded through PMC. The wound prototype cites the supplied review as REVIEW_SYNTHESIS/OTHER and explicitly retains the model caveat. It is not mislabeled as a human clinical experiment.

Current repository conventions and observed patterns

Guidance read: full CLAUDE.md, supplied AGENTS.md, extend-schema, dismech-terms, dismech-references, and the PDF skill. Relevant decisions are §§1–4, §5 (Biolink only at export), §6 and quote_role (evidence), and §12 (composite measures, quantity kinds, and context-dependent effects). The evidence model distinguishes citation integrity from appraisal; validators do not establish that a marker is a valid diagnostic criterion.

The source schema, not generated code or old examples, was inspected: dismech.yaml.

The local textual census found 106 of 3,044 disorder YAML files and 25 of 177 module YAML files mentioning senescen case-insensitively. There are 34 senescence-named pathophysiology nodes in disorders and 9 in modules. Of pathophysiology cell descriptors in those files, zero disorder descriptors and one module descriptor place that stem in preferred_term. These are textual retrieval counts, not counts of established senescence mechanisms: hits can occur in evidence, negation, notes or speculative discussions. See inventory.json and the reproducible survey.py.

The module's “essentially permanent” arrest and generalized arrest→SASP prose need qualification if reused for the review's entire scope. Its use of replicative senescence on a general arrest node also must not make all inducing routes replicative. These are findings for a future focused curation pass; no existing entry was changed.

Read-only issue searches (senescence, "cell state", all states) found relevant IPF context in #3675 and #8488. Their bodies were inspected as prior design/curation context, not treated as primary biological evidence or authorization for a new slot. The searches were limited to 25 and 20 results respectively; they do not establish that no earlier discussion exists.

Ontology lookup findings

Queries and returned identifiers, definitions, obsolescence flags and ancestry are preserved in ontology-lookups.json. This is a dated lookup record, not a hand-authored term cache. New candidate IDs below came from OLS; existing prototype bindings were also read from repository label caches and checked by the repository term validator. OLS ancestors was queried separately from hierarchicalAncestors: the latter includes more than subclass ancestry.

Negative searches were bounded and recorded: CL queries senescent and senescence returned six contextual hits (including macrophages and primary cultured cell), none a dedicated senescent-cell class. A label/synonym query senescen* returned zero. GO searches senescence-associated secretory phenotype, SASP, and label/synonym secretory phenotype returned zero. This supports using a named SASP mechanism with separately bound constituent processes today; it is not proof of absence across every release or ontology. PATO and NCIT supplied the near-matches above, which were rejected as exact cell-state identities for positive semantic reasons.

Alternatives and tradeoffs

Concrete examples

The complete files are intentionally outside kb/. Disease is used as the current validation container; the prototype wrapper names/categories do not create new disease or module identities.

• 01-review-compatible.yaml: generic senescence with its caveat; an acute-wound senescent fibroblast producer; a distinct non-senescent recipient; and a general supportive SA-beta-gal readout. The general marker row is not claimed to be measured in the wound experiment.
• 02-ipf-compatible.yaml: actual IPF identity and lung context, from the existing Schafer et al. source. The human p16/severity observation is not silently converted into a particular lineage's cell count or a universally causal senescence mechanism.
• 03-extension-example.yaml: same review example, with state evidence attached only to the producer fibroblast. Generated reproducibly from example 01 and the overlay.
• 04-immune-evasion-compatible.yaml: HLA-E display by a senescent dermal fibroblast and inhibition of an NK response are separate nodes. The source typo is retained only in the review quotation; primary evidence establishes the corrected ligand identity.

The IPF example deliberately leaves the transcript biomarker term unbound. Inspection of Schafer's Results (first paragraph, Figure 1a) showed that its severity association used expression profiling; immunohistochemistry separately localized protein. Binding this transcript measurement to the more specific p16 protein term would still be the wrong kind of entity. This illustrates why finding a better-looking identifier is not sufficient.

Representative current-schema identity/context fragment from example 01:

cell_types:
- preferred_term: skin fibroblast
  term:
    id: CL:0002620
    label: skin fibroblast
  description: The senescent producer population in this node.
  located_in:
    preferred_term: skin
    term:
      id: UBERON:0002097
      label: skin of body

The senescence assertion and exact source quote sit on the named mechanism node; the CL term is not relabeled as a new ontology class. In a disease entry, locations also supplies the existing node-level location used by downstream tools. Redundant locations must agree; descriptor-level location alone is not promised to export.

Representative proposed-only fragment:

cell_types:
- preferred_term: skin fibroblast
  term:
    id: CL:0002620
    label: skin fibroblast
  cell_states:
  - name: SENESCENT
    description: Senescent producer fibroblasts in the acute wound context.
    evidence:
    - reference: PMID:42518598
      supports: SUPPORT
      quote_role: REVIEW_SYNTHESIS
      evidence_source: OTHER
      snippet: >-
        In acute cutaneous wounds, senescent fibroblasts secrete PDGF-AA to promote
        differentiation of non-senescent fibroblasts into myofibroblasts, which are
        essential in driving wound contraction and collagen production
      explanation: Page 3; review synthesis of the Demaria mouse study.

SENESCENT is an experimental authoring enum value, not a new ontology identifier. There is deliberately no meaning: GO:0090398 or exact mapping to PATO:0001487. The optional biological_processes slot on the candidate state object would name a characterizing process, not change the state's ontological type to a process. The prototype does not need that optional binding to express the paper's assertion.

Population rules if a state extension is pursued

1. Populate only when the cited passage identifies that cell population as senescent in the recorded context. Donor age, elevated p16, SA-beta-gal, PD-1, loss of proliferation, or a computational senescence score alone is insufficient for automatic assignment. Preserve weaker “senescence-like” claims as observations.
2. Record state evidence at that scope. A node's general evidence does not prove every participant is senescent. Do not attach the producer's state to recipients or effectors.
3. Preserve specimen/model, species, induction stimulus, comparator, timing and tissue in existing model fields and precise prose. A human-derived culture is IN_VITRO; a mouse result is MODEL_ORGANISM; source review synthesis remains distinguishable through quote_role. The enum is not a certainty grade.
4. Use marker readouts for observed analytes/activity and causal edges for mechanisms. p16/p21 as regulators, their measured expression, and marker-positive-cell abundance are different claims. A gene binding is not a protein assay; a protein binding is not an activity measurement. Do not put biomarker genes in disease genetic solely because they are expressed.
5. No state entry means unasserted, not non-senescent. This pilot has no negative state, time interval, transition or mixed-population fraction formalism. Separate records/nodes and prose remain necessary where these distinctions matter.
6. Do not infer disease-wide prevalence from a localized population, or a general clinical benefit from an experimental clearance intervention. Treat senolysis, SASP modulation and restoration of surveillance as distinct interventions.
7. Stage an additive pilot; do not bulk-migrate the textual census. Manually evaluate wound fibroblasts, IPF AT2 versus mesenchymal cells, chondrocytes, immune-cell senescence/exhaustion, and the Arts senescence-like counterexample. Record rejection of an unsupported state as a successful curation outcome.

Adoption would require wiring the new slot into rendering, graph/export handling, term traversal, reference validation and QC, with a decision-register entry and history record. The schema skill requires maintainer approval for merging a major schema change; it does not block this local experiment. No merge is requested here.

Queries, evidence boundaries and remaining decisions

Current-schema queries can find GO-bound senescence nodes, their CL identities, locations and module conformance. They must return an ambiguous result for a node with several cells and several processes unless the assertion is otherwise scoped. Searching the literal senescence label is useful retrieval, not automated evidence.

The proposed structure can answer “which cell populations are explicitly described as senescent, in which mechanism contexts, supported by which sources?” without parsing cell labels. It cannot yet answer “what fraction was senescent at day 7?” or “which jointly assayed markers established that state in the same cells?” A future observation model may be needed for those questions; the paper does not settle it.

The current KGX implementation emits cell participation and biological-process associations independently at disease scope. cell_type_to_edge does not consume cell_states or descriptor location; extract_nodes can use preferred_term as the name on a shared CL identifier. The export probe records these behaviors. Therefore passing a schema test is not end-to-end preservation of cell-state meaning.

Open decisions and uncertainties:

• Whether to request a more specific PATO quality/CL class, propose a GO definition revision, or keep an explicitly unmapped curated state assertion. Any request must address context-dependent arrest stability and not equate cell aging with senescence.
• Whether the immediate need is a cell–process relation, a persistent cell state, or a source-specific observation. These require different semantics.
• How to record “senescence-like,” assay-discordant populations, escape history and postmitotic contexts. This mini-review supplies no universal adjudication rule.
• Which SASP effects are demonstrated in a given disease, model and stage. Its general account does not establish every edge in every conforming disorder.
• How to separate induction rate, clearance rate, absolute burden, fraction positive, and per-cell expression. No new numerical fields are justified by this paper alone.
• Whether a future state marker panel is a set of independent observations or a computed classifier. The decision register's composite-measure gap remains open.
• Whether typed cell-state qualifiers must survive KGX and other exports as contextual assertions; changing a global CL node's name is insufficient and potentially misleading.

See VALIDATION.md for exact commands, results, limits, and local artifact paths. Production schema/data changes are not necessary to use the minimal patterns shown here; a focused future proposal can be reviewed against the explicit query and evidence limitations above.