# Celiac Dataset Tiering Rationale

32 human celiac GEO datasets; 18 relevant, 14 excluded (see inventory `reason` column).

## Tier 1 — bulk case-control (cross-study meta-analysis inputs), n=8
Whole-mucosa duodenal biopsies, active CeD vs control. Primary meta-analysis pool:
- **GSE131705** (n=54, RNA-seq) — epithelial proliferation / metabolic suppression signature
- **GSE106260** (n=52, array) — childhood CeD, epithelial-cell focus
- **GSE164883** (n=48, array) — duodenal transcriptomic heterogeneity
- **GSE123649** (n=34, RNA-seq) — tissue-resident immunity reshaping
- **GSE92472** (n=15, RNA-seq), **GSE146190** (n=11, RNA-seq) — smaller confirmatory
- **GSE4592 / GDS2042** (n=10, array) — same CTL→NK-reprogramming study (one biological dataset; use once)

Effective distinct studies for pooling: ~6-7. Mixed platform (RNA-seq + array) is a strength
for random-effects meta-analysis — platform is a between-study variance source DL pooling handles.

## Tier 2 — compartment resolution (villous-destruction axis), n=5
- **GSE277276** (scRNA-seq, n=34) — largest celiac duodenal single-cell; stromal + epithelial reprogramming — primary Stage-6 dataset
- **GSE315138** (scRNA-seq, n=8), **GSE282570** (scRNA-seq, n=6) — confirmatory single-cell
- **GSE102991** (sorted IECs), **GSE102993** (sorted IELs) — direct epithelial vs intraepithelial-lymphocyte programs; the two compartments that execute villous destruction

## Tier 3 — complementary contrasts, n=5
- **GSE145358** (GFD-treated mucosa) — covert residual damage on diet; the "accidental exposure" analog
- **GSE87629** (6-wk gluten challenge, blood) — temporal kinetics
- **GSE33078** (refractory CeD type II) — refractory population (Tier-2 target group)
- **GSE273155** (potential CeD progression), **GSE112102** (at-risk first-degree relatives)

## Excluded (14)
6 in-vitro organoid/cell-line challenge, 3 CVID/environmental-enteropathy (celiac only comparator),
3 sorted/clonal T or B cells (not whole mucosa), 1 mouse model, 1 peripheral blood.