Before you start, confirm the input route, comparison direction and completed checks. A file named deseq2 or a shared results schema does not establish which statistical engine ran.
1. Open the result tables#
Open Figures and tables and inspect the full differential-expression table before selected gene lists. Read log2FoldChange and adjusted p-values alongside the documented direction. The up/down lists apply both alpha and |log2FC| filters.
Where outputs live#
| Setting | What it changes | When to change it / example |
|---|---|---|
| results/counts/ | Gene-count matrices and applicable assignment summaries. | Absent for results-only imports; a normalized-expression table is a different quantity. |
| results/deseq2/ | Common DE outputs and route-specific normalized-expression/diagnostic tables. | Check recorded engine and input route. |
| results/enrichment/ | GO, KEGG, GSEA and configured custom-set outputs; annotation-transfer outputs, for organisms without a curated package, under results/enrichment/transfer/. | Read resource status and mapped universe with a result. |
| results/figures/ | PNG and SVG figure files. | Retain vector files for resizing and inspect exported labels. |
| results/reports/ | Self-contained HTML results report, summaries and provenance exports. | Open the stored main report when available; the cross-study report is a separate action. |
| config/, logs/, checks/ | Run configuration, execution evidence and validation records. | Preserve these alongside results when archiving a study. |
2. Read each figure for its intended purpose#
A volcano combines effect size and adjusted-p evidence. The published baseline uses unshrunken log2 fold change on its x axis. PCA, sample distance and correlation summarize sample structure when an expression matrix exists; they do not independently test the treatment contrast. Heatmap row z-scores show relative within-gene patterns, not absolute expression differences between genes.
An MA plot combines mean expression and effect size; the required mean must exist. RNA-seq-specific dispersion, Cook’s-distance and library diagnostics do not apply to every route. Imported-results projects cannot supply matrix-based PCA or heatmaps.
3. Change display settings deliberately#
| Setting | What it changes | When to change it / example |
|---|---|---|
| Volcano labels: 15; heatmap genes: 30 | Changes how many selected features are displayed. | Illustration: labels 15 → 30 adds annotations, not new DE genes. |
| Volcano y scale: cap / full / sqrt compression | Changes display of very small adjusted p-values. Cap mode marks off-scale values. | Use full height to inspect the tail, or cap to retain readability; none changes the underlying p-values. |
| PCA variable genes: 500 | Changes the genes used for the displayed PCA calculation. | This is not merely cosmetic: a different feature subset can change the PCA view. |
| Heatmap z-limit: ±2.5 | Clamps the displayed standardized colour scale. | Extreme values share a colour beyond the limit; inspect the table for exact values. |
| Show sample labels | Hides or shows sample text while retaining grouping cues. | Useful for crowded studies; keep an accompanying sample key. |
| Figure size: 6 × 5 inches; PNG 300 DPI | Changes physical export size and raster resolution. | Increase size for a crowded panel; DPI does not increase statistical information. |
| Base font 12; point size 2.5; opacity 0.55 | Changes visual readability and overlap. | Enlarge text for a slide; do not interpret a larger point as a larger effect. |
| Enrichment terms: 15 | Changes the number of terms shown. | A shortened dotplot is not the complete enrichment table. |
Leave Font family at (default serif) to request Times New Roman from the R renderer. If that family is unavailable, the renderer logs the named installed serif it uses instead; if no serif fallback exists, figure rendering stops. A family selected explicitly in the GUI should also be available to the analysis backend. Inspect the exported PNG and SVG to confirm the actual typeface. Volcano ranked-key text and heatmap/correlation guides now have more clearance; these display adjustments do not change model values.
Use the Style tab, click Regenerate figures, then Refresh figures. Vector (SVG) previews a scalable export. Figure regeneration does not rerun alignment or the primary DE model, although choices such as the PCA feature count change the displayed derived calculation.
4. Generate and inspect reports#
On Reports, open the stored main results report when it exists; Generate Reports is available for a project without one. A stored cross-study report has its own direct action, even when the current option is off. The main single-file HTML opens with an index of its sections — recorded settings, study design, figures, which genes changed, functional enrichment, sanity checks, runtime, software and provenance, and a glossary — and carries sortable tables, definitions shown on hover, focus or tap, and a figure viewer with Fit, Actual size and percentage zoom. Sorting and zooming change the display only; the saved tables hold the complete results.
Advanced: engine parity, ranking and provenance
The count engines share one results schema but not one feature set. ncbi_geneid, pca_coordinates.csv with its covariate screen, and the padj_lfc_ge_threshold companion column are written by DESeq2, edgeR and limma-voom alike; the companion column uses each method’s own interval test (DESeq2 greaterAbs, edgeR glmTreat, limma and limma-voom treat) at the configured fold-change threshold, is informational and never decides which genes are called. lfcSE carries a value on DESeq2, limma-voom and microarray limma; edgeR quasi-likelihood reports no per-gene standard error and leaves the column empty. unchanged_genes.csv and check 13 are written on the DESeq2 route only. The recorded fold-change threshold test is named in results/reports/sessionInfo.txt.
Locally fitted models export a signed model statistic for preranked use, with a documented fallback when no finite statistic is available. Imported results use confirmed log2FoldChange. Tools/references and design exports describe the route that actually ran, including installed environment specification and input provenance rather than attributing inactive defaults to the analysis. The matrix-based Wilcoxon sensitivity check is diagnostic only: when groups are small, exact p-values are discrete and power is limited, so its warning supports rank-concordance review and does not change the differential-expression calls.
Terms on this page: PCA · Provenance.