Multi-Cohort Blood Transcriptomic Meta-Analysis of Active Pulmonary Tuberculosis: Reproducibility of a 50-Gene Interferon Signature and Benchmarking of a 4-Gene Classifier

Background: Most blood transcriptomic signatures for tuberculosis (TB) come from a single population or a single platform. Few have been built by meta-analysis and benchmarked against published signatures. Methods: We meta-analysed three UK blood transcriptomic datasets (two Illumina microarray, one RNA-seq; 247 samples: 87 active pulmonary TB (PTB), 160 healthy controls (HC)). GSE107991 reanalyses the Berry test set, and two discovery datasets overlap Sweeney's discovery data, so the datasets are not fully independent. Genes were retained if they replicated in all three studies (FDR < 0.05, same direction) and passed random-effects meta-analysis (|logFC| > 1.5, FDR < 0.01). Performance was estimated in GSE28623 (The Gambia, n = 83) and GSE56153 (Indonesia, n = 36); direction was checked in GSE34608 (Germany, n = 26) and GSE81746 (India, n = 6). A LASSO and random-forest classifier was compared with the Berry 393-gene and Sweeney 3-gene signatures. Immune composition was assessed with MCPcounter and CIBERSORTx. Results: We identified a 50-gene core signature dominated by interferon-stimulated genes. All 50 were upregulated and reached FDR < 0.05 in every dataset. ssGSEA gave an AUC of 0.830 (95% CI 0.743-0.917) in GSE28623 and 0.941 (0.868-1.000) in GSE56153. Direction agreement was 95.7% in Germany and 95.8% in India. PTB versus latent TB infection gave 0.848 (0.756-0.940). LASSO selected four genes (BATF2, ANKRD22, GBP5 and EPSTI1). The frozen classifier gave an external AUC of 0.864 (0.785-0.944). The Sweeney 3-gene score, re-tested in one of its validation cohorts, reached 0.912 and exceeded the 50-gene score (47/50 genes; Holm p = 0.0234). Berry 393-gene scoring gave 0.910 with directions reconstructed from GSE19439 (Holm p = 0.0071 versus the 50-gene score). The signed score is a reconstruction, not Berry's published score. MCPcounter showed myeloid expansion and lymphoid suppression; CIBERSORTx recovered the former but not the broader lymphoid suppression. Conclusions: The 50-gene signature was consistent across three overlapping UK discovery datasets. External evaluation was partial, comprising AUCs for 47 of the 50 genes in two cohorts and direction concordance in two smaller cohorts. This is a benchmarking study, not a claim of superiority: the Sweeney 3-gene and signed Berry scores were higher, and the 4-gene classifier did not differ significantly from Sweeney. The classifier warrants prospective validation. The Berry and Sweeney comparisons are not independent.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-20
DOI
https://doi.org/10.64898/2026.11.19.767834
Primary Topic
Tuberculosis Research and Epidemiology
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Multi-Cohort Blood Transcriptomic Meta-Analysis of Active Pulmonary Tuberculosis: Reproducibility of a 50-Gene Interferon Signature and Benchmarking of a 4-Gene Classifier

Farah Tahir, Usama Hassan, Tooba Kanwal, Armeela Bibi et al.
bioRxiv (Cold Spring Harbor Laboratory)
Tuberculosis Research and Epidemiology
preprint

Multi-Cohort Blood Transcriptomic Meta-Analysis of Active Pulmonary Tuberculosis: Reproducibility of a 50-Gene Interferon Signature and Benchmarking of a 4-Gene Classifier

Farah Tahir, Usama Hassan, Tooba Kanwal, Armeela Bibi, Warda Jamal
preprint en

Abstract

Background: Most blood transcriptomic signatures for tuberculosis (TB) come from a single population or a single platform. Few have been built by meta-analysis and benchmarked against published signatures. Methods: We meta-analysed three UK blood transcriptomic datasets (two Illumina microarray, one RNA-seq; 247 samples: 87 active pulmonary TB (PTB), 160 healthy controls (HC)). GSE107991 reanalyses the Berry test set, and two discovery datasets overlap Sweeney's discovery data, so the datasets are not fully independent. Genes were retained if they replicated in all three studies (FDR < 0.05, same direction) and passed random-effects meta-analysis (|logFC| > 1.5, FDR < 0.01). Performance was estimated in GSE28623 (The Gambia, n = 83) and GSE56153 (Indonesia, n = 36); direction was checked in GSE34608 (Germany, n = 26) and GSE81746 (India, n = 6). A LASSO and random-forest classifier was compared with the Berry 393-gene and Sweeney 3-gene signatures. Immune composition was assessed with MCPcounter and CIBERSORTx. Results: We identified a 50-gene core signature dominated by interferon-stimulated genes. All 50 were upregulated and reached FDR < 0.05 in every dataset. ssGSEA gave an AUC of 0.830 (95% CI 0.743-0.917) in GSE28623 and 0.941 (0.868-1.000) in GSE56153. Direction agreement was 95.7% in Germany and 95.8% in India. PTB versus latent TB infection gave 0.848 (0.756-0.940). LASSO selected four genes (BATF2, ANKRD22, GBP5 and EPSTI1). The frozen classifier gave an external AUC of 0.864 (0.785-0.944). The Sweeney 3-gene score, re-tested in one of its validation cohorts, reached 0.912 and exceeded the 50-gene score (47/50 genes; Holm p = 0.0234). Berry 393-gene scoring gave 0.910 with directions reconstructed from GSE19439 (Holm p = 0.0071 versus the 50-gene score). The signed score is a reconstruction, not Berry's published score. MCPcounter showed myeloid expansion and lymphoid suppression; CIBERSORTx recovered the former but not the broader lymphoid suppression. Conclusions: The 50-gene signature was consistent across three overlapping UK discovery datasets. External evaluation was partial, comprising AUCs for 47 of the 50 genes in two cohorts and direction concordance in two smaller cohorts. This is a benchmarking study, not a claim of superiority: the Sweeney 3-gene and signed Berry scores were higher, and the 4-gene classifier did not differ significantly from Sweeney. The classifier warrants prospective validation. The Berry and Sweeney comparisons are not independent.

bioRxiv (Cold Spring Harbor Laboratory)
Riphah International University (PK), University of Haripur (PK)
Tuberculosis Research and Epidemiology
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon