Donor-level cross-dataset assessment of a prespecified NF-κB-related score in human bone marrow aging — reproducibility archive

This archive contains the complete, auditable analysis for the manuscript "Donor-level cross-dataset assessment of a prespecified NF-κB-related score in human bone marrow aging" (v0.10). Study summary. We performed a donor-level, cross-dataset assessment of a fixed 30-gene NF-κB-related score in three small, publicly available human bone-marrow transcriptomic datasets (GSE115348, GSE139073, GSE94736). In the primary dataset, the score was positively associated with age in hematopoietic progenitor cells (β=0.036 per 10 years, p=0.0075, FDR=0.0447 under log1p(CPM) with an unweighted mean score). This association was donor-robust, gene-robust, and permutation-consistent, but it did not survive an alternative library-size normalization (upper-quartile), was not significant under the most conservative two-cluster framing of the data (Welch p=0.075), and did not agree in direction with the external estimates (GSE94736 was significantly opposite, p=0.0066). The study neither establishes nor excludes a pathway-level NF-κB transcriptional signature in bone-marrow aging; its contribution is an auditable donor-level framework and a demonstration that a plausible association dissolves under most reasonable alternative analyses. Archive contents. Fixed gene set (not formally preregistered; provenance documented); analysis scripts (donor-level age association, robustness, age-structure audit, external re-scoring under an identical construction, two-cluster/inference reconciliation, construct-validity controls, figure generation); all derived result tables (Supplementary Tables S1–S9); primary donor-level gene effects and pathway scores; and dataset-specific audit records. A SHA-256 manifest is included. Reproduction. All reported statistics were verified to reproduce exactly from the included scripts. To re-run from scratch, download the public datasets (GSE115348, GSE139073, GSE94736 from GEO; HRA003258 from GSA-Human) and place the count matrices as documented in each script header. Raw data are not redistributed here. Corresponding author: Xing Lu, MD, Department of Orthopedic Surgery, Yangquan First People's Hospital, Shanxi, China.