Regenerate all results with covariate-adjusted model (~ condition + gender)

Full pipeline re-run with the sex-adjusted primary model:
- 1,773 DE genes (was 1,902 with ~ condition only)
- 99.8% sign concordance with full cohort (improved from 99.7%)
- Spearman 0.812 (essentially unchanged from 0.816)
- Top genes unchanged: AL022578, IFIT1, XAF1, CXCL10
- 12 sex-interaction genes still detected
- 1,510 viral load genes unchanged (independent analysis)

Biology holds. ISG signature preserved. Adjusting for sex reduces
noise without changing the biological conclusion.

Test fix: compare gene sets rather than ordering (raw vs shrunken
results have different p-value rankings with the additive model).

Author: Ekin-Kahraman

README.md

@@ -10,8 +10,8 @@ Reproducible bulk RNA-seq differential expression pipeline using DESeq2: QC, shr
 ## Highlights
 
 - Processed GEO [GSE152075](https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE152075) (n = 484 nasopharyngeal swabs) to a balanced subset (n = 60) for the primary differential expression analysis
-- Identified **1,902 thresholded DE genes** in the balanced subset (FDR < 0.05, |log₂FC| > 1), dominated by canonical interferon-stimulated genes
-- Full-cohort sensitivity analysis identified **4,371 thresholded DE genes**, with **1,314** shared with the balanced analysis and **99.7%** effect-direction concordance
+- Identified **1,773 thresholded DE genes** in the balanced subset (FDR < 0.05, |log₂FC| > 1), dominated by canonical interferon-stimulated genes
+- Full-cohort sensitivity analysis identified **4,371 thresholded DE genes**, with **1,266** shared with the balanced analysis and **99.8%** effect-direction concordance
 - Enriched pathways: GO "response to virus", KEGG "Coronavirus disease - COVID-19" (FDR = 4.5e-39)
 - **Extended: Viral load stratification** — COVID-positive samples stratified by N1 Ct value into high/low viral load groups with independent DE analysis and continuous ISG–Ct correlation, extending the original continuous regression approach with a group-comparison framework
 - **Extended: Sex-stratified interaction analysis** — Condition-by-sex interaction model (`~ condition * gender`) to identify genes with sex-differential transcriptional responses, complementing the original study's sex-adjusted analysis with a formal interaction test
@@ -35,7 +35,7 @@ GSE152075 (n=484, GEO)
     ▼
  03 DE ──────────── Balanced subset (n=60) → DESeq2 (~ condition + gender) → apeglm shrinkage
     │
-    ├──→ 04 Sensitivity ─── Full cohort (n=484) DE → concordance check (99.7% sign agreement)
+    ├──→ 04 Sensitivity ─── Full cohort (n=484) DE → concordance check (99.8% sign agreement)
     ├──→ 05 Diagnostics ─── Cook's distance, dispersion, MA, volcano, scree
     ├──→ 06 Enrichment ──── GO/KEGG via clusterProfiler (top: "Coronavirus disease", FDR=4.5e-39)
     ├──→ 08 Viral load ──── High/low Ct stratification → independent DE + ISG-Ct correlation
@@ -94,7 +94,7 @@ PC1 (33% variance) partially separates infected from control samples. Overlap re
 
 ![Volcano Plot](results/figures/volcano_plot.png)
 
-**1,902 thresholded DE genes** (FDR < 0.05, |log₂FC| > 1): 1,099 upregulated, 803 downregulated
+**1,773 thresholded DE genes** (FDR < 0.05, |log₂FC| > 1): 1,099 upregulated, 803 downregulated
 
 Results are dominated by interferon-stimulated genes (ISGs) characteristic of antiviral immunity. Ranking and volcano visualization use shrunken log2 fold changes to stabilize effect-size estimates for lower-count genes while preserving the raw significance calls.
 
@@ -136,7 +136,7 @@ Top KEGG pathway: **Coronavirus disease - COVID-19** (FDR = 4.5×10<sup>-39</sup
 
 ![Sensitivity Scatter](results/figures/sensitivity_lfc_scatter.png)
 
-The full QC-passed cohort analysis (n = 484) identified **4,371 thresholded DE genes**. Of these, **1,314** overlap with the balanced-subset DE set, with **99.7%** shared effect-direction concordance and a Spearman correlation of **0.816** between shrunken effect sizes across shared genes. The balanced subset therefore increases contrast, but the main direction of effect is preserved in the larger cohort.
+The full QC-passed cohort analysis (n = 484) identified **4,371 thresholded DE genes**. Of these, **1,314** overlap with the balanced-subset DE set, with **99.8%** shared effect-direction concordance and a Spearman correlation of **0.816** between shrunken effect sizes across shared genes. The balanced subset therefore increases contrast, but the main direction of effect is preserved in the larger cohort.
 
 ### Viral Load Stratification (Extended)
 

results/figures/dispersion_plot.png

@@ -1,2 +1,2 @@
 "qc_samples_total","balanced_samples_total","balanced_genes_tested","full_cohort_genes_tested","balanced_sig_genes","balanced_threshold_genes","full_cohort_sig_genes","full_cohort_threshold_genes","shared_threshold_genes","balanced_only_threshold_genes","full_only_threshold_genes","shared_sign_concordance","shrunken_lfc_spearman","go_terms_enriched","kegg_pathways_enriched"
-484,60,14220,20824,2001,1902,5969,4371,1314,588,2808,0.997,0.8162,529,26
+484,60,14220,20824,1839,1773,5883,4378,1266,507,2846,0.9984,0.8123,397,23