Whole-blood GBP5 probe intensity in active tuberculosis, latent infection and pulmonary comparators: an exploratory secondary analysis of three public microarray datasets

This archive contains every script, extracted value and derived table needed to reproduce each number reported in the manuscript "Whole-blood GBP5 probe intensity in active tuberculosis, latent infection and pulmonary comparators: an exploratory secondary analysis of three public microarray datasets", together with the code that generated the figures. It contains no primary data and no identifiable information. Background. Whole-blood GBP5 transcript abundance is a widely cited candidate marker for distinguishing active tuberculosis (TB) from latent TB infection (LTBI). The manuscript re-analyses three publicly deposited whole-blood microarray series to test how far that signal survives comparison with clinically realistic alternative diagnoses, adjustment for blood cell composition, and probe-level annotation. Data sources. All analyses use only the processed values and sample metadata already deposited in the NCBI Gene Expression Omnibus: • GSE28623 (The Gambia; GPL4133 Agilent-014850 4x44K) — 46 active TB, 25 LTBI, 37 uninfected donors • GSE19491 (London and South Africa; GPL6947 Illumina HumanHT-12 V3) — TB and LTBI records • GSE42834 (London; GPL10558 Illumina HumanHT-12 V4) — 27 TB records and pulmonary comparator groups download_data.sh re-downloads the six required files (three series matrices and three platform annotation files) directly from the NCBI FTP server. They are not redistributed here because they remain under the terms of the original deposit. Contents. Analysis code • common.py — shared loaders; applies the same transformation rules to each series • meta_only.py — streams sample metadata without loading full matrices (memory constraint for the two large Illumina series) • download_data.sh — re-downloads the six source files • gse28623_analyses.py — primary comparison, sex adjustment, GBP paralogue panel, module score, label-noise simulation • gse19491_analyses.py — reproduction, participant de-duplication, paralogue panel • gse19491_spectrum.py — smear/culture-stratified spectrum analysis • gse42834_analyses.py — comparison against pneumonia, lung cancer and sarcoidosis groups; subject-duplication check • essential_analyses.py — sample-identity audit, adjusted analysis, cell-composition proxies, probe-annotation verification • composition_grid.py, composition_robust.py — composition model grid, collinearity diagnostics and alternative marker sets • sarcoid_dedup.py — active-sarcoidosis duplication check (all 33 records versus one record per subject) • table_numbers.py — computes every value quoted in Tables 2 and 3 • make_figures.py — builds Figure 1 and Figure 2 and the record-level supplement Results (JSON: machine-readable output of each script) • gse28623_results.json, gse19491_results.json, gse19491_spectrum.json, gse42834_results.json • essential_results.json, composition_grid.json, composition_robust.json, sarcoid_dedup.json • contrasts_5000.json, table_numbers.json Derived tables (CSV) • record_level_gbp5.csv — one row per selected record, with group, series and probe intensity • gse28623_record_level.csv, gse28623_gbp_probes.csv • gse19491_selected_records.csv, gse19491_aucs.csv, gse19491_titles_parsed.csv, gse19491_tb_smear_culture.csv • gse42834_comparators.csv, gse42834_gbp_panel.csv, gse42834_identity_audit.csv, gse42834_wholeblood_meta.csv • probe_annotation_check.csv — probe-to-gene and accession mapping for the GBP family on all three platforms Reproduction. From the archive root: bash download_data.sh cd analysis python gse28623_analyses.py python gse19491_analyses.py python gse19491_spectrum.py python gse42834_analyses.py python essential_analyses.py python composition_grid.py python composition_robust.py python sarcoid_dedup.py python table_numbers.py python make_figures.py Each script writes its own JSON/CSV output; the manuscript quotes only values that appear in these files. sarcoid_dedup.py and composition_grid.py run from the extracted tables and need no download; the remaining scripts read the six files fetched by download_data.sh. The label-noise simulation is seeded (20261103) and reproduces exactly. Software environment. Python 3.13 (numpy 2.3, pandas 2.2, scipy 1.17, matplotlib 3.11). No other dependencies and no proprietary software. Scope and limitations, stated explicitly. The archive documents what the deposited data can and cannot support. HIV serostatus, treatment timing, age for GSE42834, and prospective assay performance are not present in the deposits and were not imputed or inferred. Cell composition is represented by lineage marker-gene scores, not by deconvolution or sorted cells; several marker transcripts are themselves interferon-inducible, so the composition analysis bounds rather than isolates that contribution. Comparison with the same deposited TB records across comparator groups is exploratory and post hoc. Every adjustment performed here is reported with the covariate set named. Ethics. The study used only de-identified, publicly available data. No participants were recruited, no identifiable information was accessed, and no biological material was used; ethical approval was therefore not required. Ethical approval and informed consent for each original study were obtained by the original investigators and are described in the source publications (Maertzdorf et al. 2011; Berry et al. 2010; Bloom et al. 2013). Licence. Code and derived tables are released under CC BY 4.0. The underlying expression data remain subject to the terms of their original GEO deposit. Reproducible analysis archive for "Whole-blood GBP5 probe intensity in active tuberculosis, latent infection and pulmonary comparators: an exploratory secondary analysis of three public microarray datasets" GBP5, tuberculosis, latent tuberculosis infection, blood cell composition, gene expression microarray, diagnostic accuracy, reproducible analysis, secondary analysis Reproducible analysis archive for a secondary analysis of three public whole-blood microarray series (GEO: GSE28623, GSE19491, GSE42834) examining whole-blood GBP5 probe intensity in active tuberculosis, latent infection and pulmonary comparators. Contains the analysis scripts, machine-readable results, derived record-level tables and probe-annotation checks behind every reported statistic, plus the code that generated the figures. No primary or identifiable data are included; the source data are re-downloaded from NCBI GEO by the included script.