Research use only: This site is an integrated query resource for publicly available chronic rhinosinusitis with nasal polyps (CRSwNP) transcriptomic data, for research reference only. It does not provide clinical diagnosis, disease probability estimates, or treatment recommendations. All values come from uniform reanalysis of public data; known limitations of each cohort are annotated alongside the data.
Workflow
Census & curation: each of the 66 human CRS / nasal-polyp GEO series (census of 1 October 2026) was individually verified (tissue / grouping / platform / sample size) and graded core / pending / excluded with per-series evidence notes; endotype labels come only from deposited metadata, never inferred.
Uniform bulk reanalysis: independent differential analysis per cohort (counts → DESeq2; continuous values → limma; paired designs → paired models), preceded by a three-way scale check; probes collapsed to symbols by min-p; cross-platform aggregation at the results level only — expression matrices are never merged.
Unified single-cell annotation: Leiden clustering + canonical marker scoring to a unified 25-class vocabulary (scored on the full gene matrix, not an HVG subset); positive-control assertions (housekeeping / epithelial / immune markers) must pass before results are released; differential statistics at donor level (pseudobulk).
Spatial: GeoMx organized by compartment; with 6-vs-4 donors no gene reaches FDR significance, so the spatial layer serves only as a compartment-localization/display layer, not independent evidence; descriptive values are shown with fixed annotations of low power and tissue confounding.
Website: precomputed results rendered as a static viewer; known limitations of each dataset are annotated next to the data.
Analysis layer: pathway enrichment (over-representation on up/down genes separately + GSEA on powered contrasts), per-sample ssGSEA program scores (13 immune/epithelial programs), MCP-counter-style immune deconvolution (labeled abundance estimates), and CellChat ligand–receptor analysis of the flagship single-cell dataset.
Known limitations (full list)
Endotype (eosinophilic / non-eosinophilic) labels come from deposited metadata in only 6 series, most with small samples — this site does not predict tissue eosinophil counts; it only shows program-/gene-level evidence.
Six bulk cohorts have n≤7 per group or low power (GSE72713, GSE150846, GSE174681, GSE330761, GSE36830, GSE194282) — their weight in the evidence matrix should be discounted.
GSE198950 has tissue confounding, GSE69093 a genuine weak signal, GSE189690/GSE330761 no healthy controls — cross-cohort comparisons must be stratified, never directly merged.
GSE158277 (paired N-ERD, n=20): GEO has only a results table, no expression matrix — reanalysis was not possible; it can be added if raw data become available.
Single-cell: sorted preparations (GSE179292/GSE324445/GSE270953/GSE255544, etc.) do not represent whole tissue; GSE324445’s nominal sort disagrees with measured composition; GSE270953 uses tonsil controls; GSE276503 has only 2 controls.
Spatial: see the four spatial-layer limitations on the ‘Cohorts & datasets’ page.
Secondary analysis of public data: ethics approval for the original studies rests with those studies; this site collects no new samples or personal information.
Analysis layer: deconvolution values are abundance estimates (marker scoring), not cell counts; CellChat probabilities are small (~1e-6) and classical type-2 cytokine pairs fall below reliable single-cell detection, so no cytokine-pair claims are made.
Data sources
All data come from public GEO series (PMID and GEO links for each series are on the ‘Cohorts & datasets’ page); the curation and reanalysis reports plus this site’s build scripts are public with the code repository, so every number is reproducible.
Authors & contact
Authors: Peiliang Lei · Corresponding: Peiliang Lei (of@live.com) · Cite this resource: [to be filled]