Cross-Study Meta-Analysis
Combine findings across multiple independent datasets to identify replicated signals and distinguish robust biology from study-specific artifacts. Cortex runs an identical analytical pipeline on each dataset independently, then performs systematic cross-study comparison using concordance metrics, effect size comparisons, and heterogeneity assessment. Findings that replicate across independent experiments carry far more weight than single-study results.
Deliverables
Replicated molecular signatures
Findings that reproduce across independent datasets are flagged as high-confidence. In the IL-33 analysis, three key findings replicated across GSE282860 (cytokine stimulation) and GSE282861 (antibody blockade): IL-33 as the broadest responder, NF-κB/AP-1 as its primary signaling axis, and STAT6 restriction to IL-4/IL-13 signaling. These convergent signals from independent experimental designs (stimulation and blockade) provide the strongest possible evidence for biological mechanisms.
Cross-study concordance metrics
Quantitative assessment of agreement across studies using Pearson correlation, Spearman rank correlation, and Jaccard similarity. The IL-33/NF-κB activation finding showed high concordance (r=0.78) across studies, while the novel MEIS2 regulator finding was classified as moderate concordance (observed in study 1 only). These metrics enable transparent assessment of which findings are robust and which need additional validation.
Synergistic effects from cross-comparison
Cross-study analysis can reveal effects invisible in individual studies. The IL-4Rα blockade dataset showed 2,924 DEGs from combination therapy versus just 284 from the sum of monotherapies, a 10.3x synergistic effect. This non-additive interaction was only detectable through systematic comparison of treatment arms across the blockade study, guided by pathway knowledge from the stimulation study.
Demonstrate reproducibility across independent cohorts before committing validation resources. Focus on findings that replicate with high concordance and have clear biological interpretation.
IL-33 mechanism replication across cytokine stimulation and antibody blockade
| Study | Description | Samples | Groups | DEGs |
|---|---|---|---|---|
| GSE282860 | IL-4/IL-13/IL-33 in eosinophils & mast cells | 62 | 6 | 4,272 |
| GSE282861 | IL-33/IL-4Rα blockade in HDM airway inflammation | 43 | 7 | 2,924 |
| Finding | GSE282860 | GSE282861 | Verdict |
|---|---|---|---|
| IL-33 drives broadest transcriptional response | 1,334 MC DEGs | 2,924 DEGs (combo) | Replicated |
| NF-κB/AP-1 pathway dominates IL-33 response | NES 2.27 (MC) | NES 2.14 (combo) | Replicated |
| STAT6 drives IL-4/IL-13 but not IL-33 | Score 19.4 (eos IL-4) | Confirmed absent in IL-33 | Replicated |
Key finding: 2,924 DEGs from combination therapy vs. 284 from monotherapy sum, a 10.3x synergistic effect. This non-additive interaction was discovered only through cross-study comparison of the blockade dataset.
DUPILUMAB vs CYCLOSPORINE CONVERGENCE (GSE157194)
Dupilumab and cyclosporine show increasing convergence at higher organizational levels: gene-level correlation is only r=0.56 but module-level reaches r=0.95, meaning these mechanistically different drugs converge on the same biological programs despite different entry points.
| TF | Dupilumab delta | Cyclosporine delta | Note |
|---|---|---|---|
| STAT6 | -0.68 | -- | Dupilumab-specific suppression |
| STAT3 | -- | -0.78 | Cyclosporine-specific targeting |
| TF | Activity Score |
|---|---|
| RELA | +1.73 |
| NFKB1 | +1.62 |
| JUN | +1.59 |
| STAT3 | +1.31 |
| STAT6 | +1.02 |
| Gene | log2FC |
|---|---|
| KRT6C | +5.38 |
| SERPINB4 | +4.79 |
| S100A8 | +4.52 |
| MMP1 | +4.27 |
Methodology
Independent analysis per dataset
Each dataset is analyzed using the identical Cortex pipeline: same normalization, same differential expression method, same pathway enrichment. GSE282860 (62 samples, 6 groups) and GSE282861 (43 samples, 7 groups) were processed independently to avoid cross-contamination of results.
Gene identifier harmonization
Gene identifiers are harmonized across studies to enable direct comparison. Different platforms, annotation versions, and naming conventions are reconciled into a unified gene-level representation spanning 19,634 shared genes.
Effect size and direction comparison
For genes significant in both studies, effect sizes (log₂ fold changes) and directions are compared. Agreement in both magnitude and direction provides strong evidence for replication. Discordant results are flagged for heterogeneity analysis.
Concordance metric computation
Pearson and Spearman correlations measure global agreement across shared genes. Jaccard similarity quantifies overlap of significant gene sets. Per-finding concordance scores enable assessment at different levels of granularity.
Heterogeneity assessment
Each finding is assessed for heterogeneity across studies. High-concordance findings (like IL-33/NF-κB activation) are marked as robust. Study-specific findings (like MEIS2) are flagged as requiring independent validation before they can be considered generalizable.
Target personas
Meta-analysis lead
Run systematic cross-study comparisons with quantified concordance metrics and transparent heterogeneity assessment.
Systematic review author
Combine evidence across studies with reproducible, auditable methodology and full provenance tracking.
Evidence synthesis team
Build evidence packages grounded in replicated findings from independent datasets, not single-study results.
Explore more
Ready to see your own data analyzed?
Tell us what you're working on. We can show you what the output would look like.