Measured DMD evidence · Stage C2

DMD induction, patient replication and CRISPR correction are now connected by an auditable pathway analysis.

This release adds biological-repeat-aware Reactome tests for GSE272233 and sample-level external disease validation. It strengthens disease-context evidence and provides reference axes for a prespecified prospective candidate-validation study.

Open API summary Download Figure 8 Verify objects

Measured disease context, prediction still lockedDMD-locus perturbation and patient-muscle pathway evidence are observed. Candidate-gene response prediction, therapeutic efficacy and patient-level simulation remain unavailable.

Evidence chain

Four layers are shown together, but their inferential roles remain separate.

Inspect statistical units →

DMD induction

Engineered and patient DMD agree descriptively

GSE233606 provides an unselected gene-axis bridge between engineered DMD and a patient line. It is descriptive, not donor-level inference.

Patient replication

291 camera-supported pathways

PRJNA772047 uses five reported donor/sample pseudobulks. PRJNA1218493 and mouse GSE156497 remain directional sensitivity layers.

CRISPR correction

24 background-specific reversals

GSE272233 uses three reported biological repeats per group. No dual-supported pathway reverses across all three mutation backgrounds.

Organoid heterogeneity

Context dependence retained

Organoid evidence is shown as a heterogeneity layer and is not pooled into a universal DMD response.

NAR main figure

Disease induction → patient replication → correction → heterogeneity

Accessible SVG Source data PDF

Four-panel measured DMD evidence figure showing induction, external patient pathway agreement, CRISPR correction and organoid heterogeneity. — Figure 8 keeps descriptive, formal and sensitivity-only evidence visually distinct. The lower strip summarizes the four measured evidence layers and their inferential units.

Figure 8 long description, data table and methods

Long description. Panel A shows a positive descriptive association between the unselected engineered-DMD and patient-DMD gene axes (Spearman 0.4498 across 23,324 genes; direction agreement 68.29%). Panel B separates one formal patient pseudobulk test from two directional sensitivity datasets: 291 PRJNA772047 pathways have camera support, while 54.73% of 1,480 shared pathways keep the same direction across all three external datasets. Panel C compares three DMD-locus correction backgrounds; 24 dual-supported pathway reversals occur only in dup8–9 and none are shared by all backgrounds. Panel D shows decreasing organoid-to-disease pathway agreement across DMD1, DMD2 and DMD3 (0.6803, 0.6413 and 0.5533), preserving heterogeneity rather than pooling it away.

Accessible key-value table for Figure 8. The downloadable TSV contains all 20 plotted rows.
Panel	Evidence layer	Metric	Value	n	Inference
A	DMD induction	Unselected gene-axis Spearman	0.4498	23,324 genes	Descriptive; one line per condition
A	DMD induction	Direction agreement	68.29%	23,324 genes	Descriptive; no cell-level P values
B	Patient replication	camera-supported pathways	291	1,526 pathways	3 DMD vs 2 normal donors/samples
B	External sensitivity	Same direction in all three datasets	54.73%	1,480 pathways	Directional agreement only
C	CRISPR correction	Background-specific dual reversals	24	1,559 pathways	Observed only in dup8–9
C	Cross-background correction	Dual reversal in all backgrounds	0 · measured null	3 backgrounds	Measured absence under the frozen rule
D	Organoid heterogeneity	Direction agreement range	0.5533–0.6803	1,511 pathways	Descriptive; shared WT reference

Methods boundary. Panel A is descriptive. Panel B uses camera only for PRJNA772047 and direction-only sensitivity for the other datasets. Panel C keeps mutation backgrounds separate and uses the frozen dual-support rule. Panel D is a heterogeneity display, not a pooled universal response. Open figure-generation code.

Primary statistical readout

Strict consensus is sparse, and that result is preserved.

Dual reversal in all three correction backgrounds0 · measured null

Strict independent reversal in all three backgrounds0 · measured null

Background-specific dual-supported reversals24

Patient pathways with camera support291

The 492 fixed-effect cross-background reversals are explicitly exploratory because WT is non-isogenic and heterogeneity is substantial. They are not promoted to validated virtual-cell predictions.

Reusable objects

API, pathway tables, sample QC and code share one release identity.

GSE272233 API External validation API Sample QC API camera results RO-Crate only · open manifest External effects RO-Crate only · open manifest

Independent rebuild receipt

All 16 clean-room comparisons pass.

The path-isolated rebuild regenerated GSE233606 induction, GSE272233 correction and camera tests, three external pseudobulk pathway layers, GSE277637 heterogeneity and Figure 8. Numeric tables and metadata-independent figure pixels match the released analysis.

Open clean-room API Download report Download audit

Boundary: this is path/process-isolated computational reproduction using the same system Python/R packages, not a container-level independent-host reproduction and not biological validation.

Claim boundary

Supported: Measured DMD-locus induction/correction and sample-level external disease pathway replication with explicit small-n limitations.

Not supported: Candidate-gene response prediction in DMD cells, therapeutic efficacy, patient-level simulation and clinical decision support.