Table 3.
List of studies on ML methods based on transcriptomics technology in the differential diagnosis of LTBI and ATB
| Study | Sample size | Countries | Methodsψ | Biomarkers | ML methods | Sensitivity | Specificity | Accuracy | AUC |
|---|---|---|---|---|---|---|---|---|---|
| Lee et al. [238] | TB (n = 15), LTBI (n = 17), HCs (n = 15) | Taiwan, China | Microarray | PTPRC + ASUN + DHX29 + NEMF | Decision tree, random forest, support vector machine, Bayesian best | 97.9% with PTPRC + ASUN + DHX29 under the Bayesian model | Unknown | 97.8% | 0.979 |
| Lu et al. [239] |
Discovery cohort, TB (n = 4), LTBI (n = 4), HCs (n = 4); qPCR validation cohort, TB (n = 25), LTBI (n = 36), HCs (n = 22); additional validation cohort, TB (n = 17), LTBI (n = 19) |
China | Microarray | CXCL10, ATP10A, and TLR6 combination | Decision trees and unsupervised cluster analysis | 71% in the additional validation cohort | 89% in additional validation cohort | Unknown | Unknown |
| Wang et al. [240] | Identification cohort, ATB (n = 28), LTBI (n = 25), HCs (n = 31); validation cohort, ATB (n = 51), LTBI (n = 44), HCs (n = 35) | China | RNA-seq | TNFRSF10C, IFNG, PGM5, EBF3, A2ML1 | Decision trees and unsupervised cluster analysis | 86.2% with a combination of TNFRSF10C, EBF3, and A2ML1 | 94.9% | 87.8% | Unknown |
| Maertzdorf et al. [241] | ATB patients (n = 120), LTBI (n = 60), HCs (n = 20); external cohorts, from the Gambia (n = 75), from the Uganda (n = 62) | Africa | RT-PCR | GBP1, IFITM3, P2RY14, and ID3 | Random forest, decision tree | Using a cutoff of 0.8, Uganda: 73%, Gambia: 85%; using a cutoff of 0.6, Uganda: 87%, Gambia: 88% | Using a cutoff of 0.8, Uganda: 78%, Gambia: 76%; using a cutoff of 0.6, Uganda: 75%, Gambia: 68% | 82% in Uganda and 89% in Gambia | AUC = 0.89 in Gambia and AUC = 0.82 in Uganda |
| Bayaa et al. [242] | ATB patients (n = 141), LTBI (n = 26), HCs (n = 71) | Multiple countries | qRT-PCR | RISK6 | No | 90.9% | 88.5% | Unknown | 0.930 |
| Gong et al. [243] | ATB patients (n = 51), lung cancer (n = 30), INFLA (n = 30), HCs (n = 15) | China | qRT-PCR | SERPING1, BATF2, UBE2L6, and VAMP5 | No | 88% | 78% | Unknown | 0.840 |
ψMethods used for screening and identification of biomarkers
A2ML1 alpha-2-macroglobulin like protein 1, ASUN asunder spermatogenesis regulator, ATB active tuberculosis, ATP10A ATPase phospholipid transporting 10A, AUC area under curve, BATF2 basic leucine zipper transcription factor, CXCL10 chemokine (C-X-C motif) ligand 10, DHX29 DEAH (Asp-Glu-Ala-His) box polypeptide 29, EBF3 early B cell factor 3, GBP1 guanylate binding protein 1, HCs health controls, ID3 inhibitor of DNA binding 3, IFITM3 interferon-induced transmembrane protein 3, IFNG interferon-γ gene, INFLA patients with pneumonia, LTBI latent tuberculosis infection, ML machine learning, NEMF nuclear export mediator factor, P2RY14 UDP-glucose-specific G(i) protein-coupled P2Y receptor, RT-PCR reverse transcription-polymerase chain reaction, PGM5 phosphoglucomutase 5, PTPRC protein tyrosine phosphatase receptor type C, qRT-PCR quantitative real-time PCR, RNA-seq RNA-sequencing, RISK6 six whole blood gene transcriptomic signature, SERPING1 serpin peptidase inhibitor C1 inhibitor member 1, TNFRSF10C TNF receptor superfamily member 10C, TLR Toll-like receptor, TB tuberculosis, UBE2L6 ubiquitin-conjugating enzyme E2L6, VAMP5 vesicle-associated membrane protein 5