Abstract
Biomarkers are critical for rapid diagnosis of tuberculosis (TB) and could benefit patients with AIDS where diagnosis of TB co-infection is challenging. Meta-analysis is an approach to combine the results of the studies with standard statistical method by weighting each study with different sample size. This study aimed to use meta-analysis to integrate transcriptome datasets from different studies and screen for TB biomarkers in patients who were HIV-positive. Five datasets were subjected to meta-analysis on whole-blood transcriptomes from 640 patients infected with HIV. A total of 293 differentially expressed genes (DEGs) were identified as significant (P<0.0001) using the random effective model to integrate the statistical results from each study. DEGs were enriched in biological processes related to TB, such as "Type I interferon signaling" and "stimulatory C-type lectin receptor signaling". Eighteen DEGs had at least a two-fold change in expression between patients infected with HIV who were TB-positive and those who were TB-negative. GBP4, SERPING1, ATF3 and CDKBN3 were selected as a biomarker panel to perform multivariable logistic regression analysis on TB status and relative gene expression levels. The biomarker panel showed excellent accuracy (AUC>0.90 for HIV+TB) in clinical trial and suggests that meta-analysis is an efficient method to integrate transcriptome datasets from different studies.