Tuberculosis (TB) remains a highly contagious public health threat. Precise and prompt diagnosis and monitoring of treatment responses are urgently needed for clinics. To pursue novel and satisfied host blood-derived biomarkers, we streamlined a bioinformatic pipeline by integrating differentially expressed genes, a gene co-expression network, and short time-series analysis to mine the published transcriptomes derived from whole blood of TB patients in the GEO database, followed by validating the diagnostic performance of biomarkers in both independent datasets and blood samples of Chinese patients using quantitative real-time PCR (qRT-PCR). We found that four genes, namely UBE2L6 (Ubiquitin/ISG15-conjugating enzyme E2 L6), BATF2 (Basic leucine zipper transcriptional factor ATF-like), SERPING1 (Plasma protease C1 inhibitor), and VAMP5 (Vesicle-associated membrane protein 5), had high diagnostic value for active TB. The transcription levels of these four gene combinations can reach up to 88% sensitivity and 78% specificity (average) for the diagnosis of active TB; the highest sensitivity can achieve 100% by parallel of BATF2 and VAMP5, and the highest specificity can reach 89.5% through a combination of SERPIG1, UBE2L6, and VAMP5, which were significantly higher than 75.3% sensitivity and 69.1% specificity by T-SPOT.TB in the same patients. Quite unexpectedly, the gene set can assess the efficacy of anti-TB response and differentiate active TB from Latent TB infection. The data demonstrated these four biomarkers might have great potency and advantage over IGRAs in the diagnosis of TB.