Study · Endometrial regulons

Robustness and cross-cohort concordance of developmental regulons in endometrial carcinoma

Prespecified TCGA-UCEC discovery with CPTAC-UCEC external evaluation, target-deletion robustness and clinical sensitivity analyses.

Authors

Draga Toncheva1Vasil Sgurev2Vladimir Mitev3

  1. 1Department of Biological Sciences, Bulgarian Academy of Sciences
  2. 2Institute of Information and Communication Technologies, Bulgarian Academy of Sciences
  3. 3Helena Bioinformatics EOOD

Abstract

This study asks whether the molecular classes of endometrial carcinoma differ in developmental regulatory programmes while keeping association, target-set robustness and cross-cohort transportability as separate claims.

A fetal Müllerian epithelial module and 20 prespecified developmental transcription-factor regulons were evaluated in TCGA-UCEC. Six signals then underwent external evaluation in two independent CPTAC-UCEC strata, together with deletion, purity, molecular-subtype, grade and histology sensitivity analyses.

Read the preprintView on GitHub

Two regulons hold their prespecified direction in an independent cohort. None is presented as a validated biomarker.

01 · Effect estimates

Cross-cohort forest plot

−1.5−1.0−0.50.0+0.5+1.0+1.5GATA2C2GATA2: d −0.553, 95% CI −0.855 to −0.287, q 0.0020−0.553SOX9C2SOX9: d −0.527, 95% CI −0.805 to −0.287, q 0.0020−0.527HOXA9C2HOXA9: d −0.646, 95% CI −0.921 to −0.382, q 0.0020−0.646WT1C2WT1: d −0.591, 95% CI −0.878 to −0.349, q 0.0020−0.591PAX8C1PAX8: d +0.693, 95% CI +0.473 to +0.924, q 0.0020+0.693LHX1C1LHX1: d +0.962, 95% CI +0.734 to +1.207, q 0.0020+0.962standardized effect size d
ZeroLocked |d| = 0.50 floorValues and intervals: GitHub scientific release
This site presents cohort-level research results. It does not provide patient-specific interpretation, treatment prediction or clinical decision support.
Releasev1.0.1
DiscoveryTCGA-UCEC · n=507
External evaluationCPTAC-UCEC · n=230
Scientific commitf702249bd50d

Study overview

Six prespecified target regulons, evaluated without a composite score.

Prespecified regulons20
Canonical targets6
Primary cohort507
External cohort230

Audit

Change one documented analytical decision and inspect the effect on the published evidence.

Purity adjustment

The switch changes the fitted model shown as the current result. Both estimates were computed in the pinned GitHub release.

Published model
−0.553−0.855 to −0.287
Purity removed
−0.539−0.842 to −0.284
No purity minus primary
+0.014
Direction
Preserved
|d| ≥ 0.50
Unchanged
CI includes zero
Unchanged

Source: six-target GitHub result table.

Remove one gene

Each option is a completed leave-one-out analysis from the GitHub deletion ledger. The most disruptive deletion is listed first.

Baseline
−0.553−0.855 to −0.287
NFE2 removed
−0.495Point estimate
Change in d
+0.058
Direction
Preserved
|d| ≥ 0.50
Falls below
Rows tested
75

Source: gene leave-one-out GitHub ledger.

Effect floor

0.50
GATA2−0.553
SOX9−0.527
HOXA9−0.646
WT1−0.591
PAX8+0.693
LHX1+0.962
Targets clearing floor
6/6
Published floor
0.50
Current floor
0.50
Model
TCGA primary

The floor changes the decision gate only. Effect estimates are unchanged.

External cohort

TCGA primary
−0.553−0.855 to −0.287
Discovery
−0.689−1.409 to −0.045
Direction
Concordant
CI includes zero
No
Sample
n=95
Change in d
−0.136

Source: CPTAC per-stratum effects and fixed-effect meta-analysis.

Clinical adjustment

Base model
−0.556Point estimate
Adjusted model
−0.560Point estimate
Change in d
−0.004
Attenuation
-0.7%
Direction
Preserved
Sample
n=505

Available for the published GATA2 and SOX9 grade and histology analyses.

Six canonical targets

No ranking or composite score is used. Every verdict is derived from the pinned GitHub result tables.

02 · Separated evidence dimensions

Evidence map

Association, target-set robustness and transportability remain separate. No composite score is calculated.

TargetAssociationRobustnessTransportability
GATA2C2 Meets locked criteriaCAT2 DEPLETION No universal creditNo universal deletion credit Direction supportedExternally replicated direction
TCGA primaryd −0.553 · 95% CI −0.855 to −0.287 · q 0.0020
TCGA no purityd −0.539 · 95% CI −0.842 to −0.284
CPTAC Discoveryd −0.689 · 95% CI −1.409 to −0.045
CPTAC Confirmatoryd −0.999 · 95% CI −1.616 to −0.497
CPTAC metad −0.877 · 95% CI −1.306 to −0.449 · q 3.62e-4
Deletion ledger1/75 primary failures · worst NFE2
SOX9C2 Meets locked criteriaCAT2 DEPLETION No universal creditNo universal deletion credit Direction supportedExternally replicated direction
TCGA primaryd −0.527 · 95% CI −0.805 to −0.287 · q 0.0020
TCGA no purityd −0.509 · 95% CI −0.775 to −0.260
CPTAC Discoveryd −0.993 · 95% CI −1.735 to −0.264
CPTAC Confirmatoryd −0.641 · 95% CI −1.322 to −0.050
CPTAC metad −0.789 · 95% CI −1.254 to −0.323 · q 0.0018
Deletion ledger15/70 primary failures · worst XPO4
HOXA9C2 Meets locked criteriaCAT2 DEPLETION No universal creditNo universal deletion credit Unresolved / underpoweredevaluable: not replicated
TCGA primaryd −0.646 · 95% CI −0.921 to −0.382 · q 0.0020
TCGA no purityd −0.614 · 95% CI −0.894 to −0.354
CPTAC Discoveryd −0.550 · 95% CI −1.255 to +0.204
CPTAC Confirmatoryd −0.195 · 95% CI −0.797 to +0.377
CPTAC metad −0.333 · 95% CI −0.789 to +0.123 · q 0.1831
Deletion ledger2/19 primary failures · worst MYCN
WT1C2 Meets locked criteriaCAT2 DEPLETION No universal creditNo universal deletion credit Unresolved / underpoweredevaluable: not replicated
TCGA primaryd −0.591 · 95% CI −0.878 to −0.349 · q 0.0020
TCGA no purityd −0.598 · 95% CI −0.861 to −0.333
CPTAC Discoveryd −0.255 · 95% CI −0.794 to +0.390
CPTAC Confirmatoryd −0.115 · 95% CI −0.604 to +0.369
CPTAC metad −0.175 · 95% CI −0.547 to +0.198 · q 0.3584
Deletion ledger1/92 primary failures · worst SULF2
PAX8C1 Meets locked criteriaCAT1 ENRICHMENT Meets locked criteriaUniversal deletion credit Unresolved / underpoweredunderpowered sensitivity
TCGA primaryd +0.693 · 95% CI +0.473 to +0.924 · q 0.0020
TCGA no purityd +0.686 · 95% CI +0.471 to +0.925
CPTAC Discoveryd +0.500 · 95% CI −0.102 to +1.140
CPTAC Confirmatoryd +1.054 · 95% CI +0.508 to +1.649
CPTAC metad +0.795 · 95% CI +0.379 to +1.212 · q 5.54e-4
Deletion ledger0/23 primary failures · worst SLC3A1
LHX1C1 Meets locked criteriaCAT1 ENRICHMENT Meets locked criteriaUniversal deletion credit Blocked by vetoOpposite-direction veto
TCGA primaryd +0.962 · 95% CI +0.734 to +1.207 · q 0.0020
TCGA no purityd +0.956 · 95% CI +0.725 to +1.217
CPTAC Discoveryd −0.031 · 95% CI −0.698 to +0.656
CPTAC Confirmatoryd +1.048 · 95% CI +0.396 to +1.834
CPTAC metad +0.470 · 95% CI −0.013 to +0.953 · q 0.0845
Deletion ledger0/6 primary failures · worst HHEX

Leave-one-target-out robustness

Deletion results test dependence on individual regulon targets; they do not measure external replication.

03 · Gene leave-one-out

Deletion robustness map

Every mark is one mapped gene removed and re-estimated under the locked pointwise gate. Orange marks fall below |d| = 0.50.

GATA2

75 mapped genes
Failures
1/75
Minimum |d|
0.4950
Worst deletion
NFE2

SOX9

70 mapped genes
Failures
15/70
Minimum |d|
0.4593
Worst deletion
XPO4

HOXA9

19 mapped genes
Failures
2/19
Minimum |d|
0.4701
Worst deletion
MYCN

WT1

92 mapped genes
Failures
1/92
Minimum |d|
0.4877
Worst deletion
SULF2

PAX8

23 mapped genes
Failures
0/23
Minimum |d|
0.5767
Worst deletion
SLC3A1

LHX1

6 mapped genes
Failures
0/6
Minimum |d|
0.6999
Worst deletion
HHEX

Cohort structure

Discovery, sensitivity and external strata remain separate to avoid overstating transportability.

TCGA primary complete cases506
TCGA no-purity507
CPTAC Discovery95
CPTAC Confirmatory135

Sensitivity analyses

Subtype decomposition and clinical adjustment are rendered directly from the pinned GitHub tables.

04 · Sensitivity analyses

Subtype and clinical sensitivity

Molecular-subtype decomposition

Bootstrap intervals for the three descriptive subtype contrasts.

POLE vs NSMP
−0.68q 0.0018
MMRd vs NSMP
−0.42q 0.0026
POLE vs MMRd
−0.26q 0.1658

Grade and histology adjustment

Matched base model versus adjusted model.

histology matched
-0.7%n=505
endometrioid grade matched
17.1%n=380
BaseAdjusted

Analysis workflow

A frozen, reproducible sequence from cohort definition to external evaluation.

Analysis workflow · Step 01

Prespecification

Twenty developmental transcription-factor regulons and the broad fetal Müllerian M1 module were frozen before external evaluation.

Purpose

Define the analysis before the independent cohort was evaluated.

Method

The M1 module, 20 regulons, target definitions and planned contrasts were fixed in the study release.

Output

A fixed analysis inventory that the discovery, robustness and external-evaluation steps use without changing the original scope.

Analysis workflow · Step 02

Discovery

Cohort-level standardized effects were estimated in TCGA-UCEC primary tumours.

Purpose

Estimate molecular-class contrasts in the primary cohort.

Method

Standardized effect sizes, confidence intervals and multiplicity-adjusted results were calculated for the prespecified TCGA-UCEC analysis.

Output

Primary estimates for the six canonical targets, retained as association evidence rather than clinical effects.

Analysis workflow · Step 03

Robustness

Purity removal and leave-one-target-out analyses tested sensitivity to modelling choices and regulon membership.

Purpose

Test whether a result depends on tumour-purity adjustment or one target gene.

Method

The completed no-purity model and target-deletion runs were compared with the primary specification.

Output

Sensitivity and deletion-robustness records that remain separate from external confirmation.

Analysis workflow · Step 04

External evaluation

CPTAC-UCEC strata were combined with a fixed-effect meta-analysis; direction, uncertainty and multiplicity were retained.

Purpose

Evaluate the prespecified direction in an independent cohort.

Method

Discovery and Confirmatory CPTAC-UCEC strata were estimated separately and then combined with a fixed-effect model.

Output

External estimates, intervals and direction-based credit or veto status for each canonical target.

Analysis workflow · Step 05

Claim boundary

Association, robustness and transportability are reported separately; no causal or clinical claim is made.

Purpose

Keep the interpretation within the evidence produced by the study.

Method

Association, robustness, external direction and uncertainty remain separate dimensions without a composite score.

Output

Cohort-level research findings that are not presented as causal activity, biomarker validation or clinical decision support.

GitHub repository

Evidence materializes a pinned scientific snapshot for stable rendering and links every displayed value back to its source file.

GitHub repositoryAnalysis code, derived results and provenanceOpenGitHub release v1.0.1Versioned reproducibility releaseOpenRelease checksumsSHA-256 integrity manifestOpenArchived software DOIZenodo software recordOpenPreprintManuscript and supplementary materialsOpen

Reuse and data boundary

Original scientific materials are CC BY 4.0; software is Apache-2.0. Third-party source terms remain applicable. Controlled raw patient data are not redistributed by Evidence.

Pinned scientific commit: f702249bd50d31697e1ffcb9c54b606b39757812

Citations

Use the preprint DOI for the scientific work and the software DOI for the reproducibility record.

Research article

Toncheva D, Sgurev V, Mitev V. Robustness and cross-cohort concordance of developmental regulons in endometrial carcinoma. 2026.

doi:10.5281/zenodo.21763096

Code and reproducibility

Toncheva D, Sgurev V, Mitev V. Endometrial developmental regulons: computational and reproducibility record. v1.0.1.

doi:10.5281/zenodo.21762179