Pre-registered transcriptomic meta-analysis of pulmonary arterial hypertension (lung tissue): a priori protocol

This deposit records the a priori protocol for a pre-registered, PRISMA-compliant multi-cohort transcriptomic meta-analysis of pulmonary arterial hypertension (PAH, WHO Group 1), conducted entirely in silico on public gene-expression data (GEO). The full protocol, PRISMA-P 2015 checklist, search log, PRISMA 2020 flow, and verified dataset inventory are attached. Objective: identify a reproducible consensus differential-expression signature distinguishing PAH from control lung across independent human cohorts, and derive a compact diagnostic panel confirmed on an independent held-out cohort. Design (fully locked in advance; see the attached registration document): primary compartment = lung tissue; results-level "analyse-then-combine" synthesis (per-study differential expression, then RobustRankAggreg and metafor random-effects meta-analysis, reporting between-study heterogeneity); consensus signature defined as meta-FDR < 0.05 AND directional consistency across cohorts AND absolute pooled log2 fold change >= 0.585 (1.5-fold); dataset inclusion requires >= 6 samples per group; mechanistic layers are single-cell cell-of-origin localisation plus immune deconvolution; a diagnostic panel is selected by LASSO logistic regression and/or random-forest importance under nested cross-validation and evaluated once on a provenance-distinct held-out cohort (ROC/AUC and calibration). Included cohorts are fixed at registration (attached inventory). Discovery (7): GSE113439, GSE117261, GSE53408, GSE15197, GSE48149 (microarray) and GSE254617, GSE272776 (RNA-seq). External validation, held out from all discovery steps: GSE208592. Single-cell reference: GSE293580. Timing disclosure: the systematic search was executed on 2026-07-16 using repository metadata only. Raw and processed source files were subsequently downloaded and staged, with integrity verified by checksum only; no expression data has been normalised, quality-control inspected, or analysed, and no differential expression, meta-analysis, single-cell projection, or classifier training has been run. All analytic decisions (thresholds, cohort split, methods, held-out validation cohort) were locked before any file was downloaded. Every analytic step is pre-specified and un-run at the time of registration.