Research Use Only. KIRhub outputs are computational research artifacts. They are not validated for clinical decision-making, diagnosis, or treatment.
KIRhub 2.0· v2Sign in
Overview·Falsification
refuted

Did the developmental-origin framework hold up?

ModeledCalibrated
Question
The field assumes pediatric ependymoma 'recapitulates fetal brain development' while adult ependymoma looks more like adult brain. Does the data actually say that?
What we found
No (per v2 — pathway-space test with a real adult-brain GTEX reference). Pediatric EPN clusters with fetal in only 10.9% of samples; both pediatric and adult EPN look more adult-brain-like than fetal-like. The textbook story does not survive.
Caveat
This is a HALLMARK pathway-level test — generic cancer biology, not developmental signaling specifically. A test against KEGG/Reactome neural-development pathways could give a different answer. We're refuting the framework at one resolution, not all resolutions.
Next step
Re-run with neural-development-specific gene signatures. The empirical PFA CDK4/6 drug-ranking findings still stand — but they no longer have a textbook mechanism story.
Scroll for the data, methods, and per-sample detail.

Adult EPN falsification test

The gate for the ependymoma developmental-origin framework (the PFA CDK4/6, ZFTA-RELA, and MERTK insights depend on this).

✗ Framework refuted (v2 — pathway space, GTEX reference)

ModeledCalibrated

Both pediatric and adult EPN cluster more with adult brain (GTEX) than with fetal reference in hallmark pathway space. The developmental-origin claim is not supported by this test. the PFA CDK4/6, ZFTA-RELA, and MERTK insights lose their developmental-mechanism justification (though empirical drug-ranking findings stand on their own).

Pediatric clusters with fetal
10.9%
vs adult-brain: 47.4%
Adult clusters with fetal
15.2%
vs adult-brain: 39.4%
EPN cohorts (cross-project)
adult=33 · ped=350
medullo-epn-umap + brain-umap
References
fetal=100 · GTEX=1409
50-d hallmark GSVA
How v2 works (pathway-space test vs v1 UMAP-space)

Pathway-space (not UMAP-space) distance — cross-project compatible because both projects compute GSVA against the same hallmark library. Caveat: hallmark pathways are largely generic cancer biology (MYC_TARGETS, GLYCOLYSIS, MTORC1, ...) not developmental signaling. A 'clusters with fetal' verdict here reflects shared proliferative/stress signaling, not necessarily cell-of-origin identity. Re-running with KEGG/Reactome neural-development pathways is the next sharpening.

Feature space: 50-d HALLMARK GSVA · distance: Euclidean · k=5 · decision ratio 0.85.

⚠ Verdict ambiguous (v1 — UMAP-space, Med contrast)

ModeledCalibrated

Adult sample size is small or adult EPN also clusters with fetal at similar rates to pediatric. Sahm/Reuss 2018 ingestion (122 adult EPN) is recommended before betting big on the developmental framework.

Adult clusters with fetal
70.8%
n=24
Pediatric clusters with fetal
89.1%
n=266
EPN samples scored
370
fetal ref n=100 · Med contrast n=888

Adult vs pediatric EPN — does each cohort land on its expected reference?

Derived

Split view of the same UMAP. If the developmental-origin framework held, the right (pediatric) panel would sit inside the fetal cluster and the left (adult) panel would sit on adult-brain reference. UMAP coordinates come from the v1 medullo-epn-umap projection; EPN points are colored red and the dashed blue circle marks the fetal-reference attractor (centroid of fetal-clustering EPN samples — see caption). v2 framework verdict (pathway-space, GTEX-anchored): refuted.

Adult EPN

n=24
UMAP1UMAP2fetal reference

Pediatric EPN

n=266
UMAP1UMAP2fetal reference
EPN samplefetal reference cluster (synthetic — see caption)adult-brain (GTEX) — not available in v1 UMAP

Caveat. The v1 medullo-epn-umap projection does not include an adult-brain (GTEX) reference — that reference lives in a different UMAP (brain-umap project) and cannot be co-located here. The green adult-brain marker in the legend is therefore unavailable. The dashed blue halo is the centroid of samples the v1 algorithm assigned to clusters_with_fetal, used here as a stand-in for the underlying fetal reference cluster (whose raw points are not returned by the API). Per-point coloring uses the v1 classification because v2 does not expose per-sample classification (only aggregate counts).

UMAP positions — does adult EPN drift off the fetal manifold?

Derived
UMAP1 (medullo-epn-umap project, 2D)UMAP2

Anatomic stratification

ModeledCalibrated

Cluster-with-fetal rate by compartment × age bucket.

CompartmentAge bucketn% clusters with fetal
otheradult1770.6%
otherpediatric5685.7%
posterior fossapediatric9489.4%
spinaladult10.0%
spinalpediatric1266.7%
supratentorialadult683.3%
supratentorialpediatric10493.3%

Downstream impact on the dependent insights

  • PFA radial-glia CDK4/6: foundation status → ambiguous
  • ZFTA-RELA TAM+MET polypharmacology: foundation status → ambiguous
  • MERTK dual-compartment: foundation status → ambiguous
  • PFA1 hypoxia-wound niche: independent of this verdict (niche-based, not developmental).

How this test runs (deviations from the spec)

The original framework asked for adult-EPN distance to adult brain reference. The medullo-epn-umap project ingested today contains 100 fetal references but no adult-brain reference in the same UMAP space — and the separate brain-umap project lives in a different UMAP that we cannot directly distance-compare.

As an in-project substitute, we use medulloblastoma as the non-fetal contrast (888 samples, same UMAP). For each EPN sample we compute mean distance to its 5 nearest fetal neighbours vs its 5 nearest Med neighbours, and call it clusters_with_fetal when the fetal distance is < 0.8× the Med distance.

Verdict is conservative. Ambiguous → recommend ingesting Sahm/Reuss 2018 (122 adult EPN) for a definitive read. Adult-age threshold ≥ 18 years.

Research Use Only · dataset saifudeen2026 1.0.0 · api 0.1.0 · 17 ms
# Platform
Saifudeen, A., et al. (2026). KIRhub: a falsification-first research workbench for translational oncology. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03090-8

# Data source
McFerrin, L. G., et al. (2018). Oncoscape: a tool for interactive cancer genomics data analysis. Nature Genetics.
Arora, S., et al. (2026). A pan-pediatric brain tumor reference map: medulloblastoma and ependymoma in shared UMAP space. Neuro-Oncology.

# Datasets used in this insight
Oncoscape pediatric brain tumor compendium (n=1,358)
GTEX adult-brain reference (n=1,409)

# Provenance
insight_id: falsification