Table 4 List of studies on ML methods based on proteomics technology in the differential diagnosis of LTBI and ATB
Study | Simple size | Countries | Biomarkers | ML methods | Sensitivity | Specificity | AUC |
|---|---|---|---|---|---|---|---|
Li et al. [263] | Discovery cohort: ATB (n = 52), LTBI (n = 37), HCs (n = 27); validation cohort: ATB (n = 205), LTBI (n = 123), HCs (n = 112); | China | Rv0934, Rv1827, Rv1860, and Rv3881c | Cluster analysis | 67.3% | 91.2% | Unknown |
Li et al. [264] | Discovery cohort: ATB (n = 60), LTBI (n = 60), HCs (n = 60); validation cohort: ATB (n = 100), LTBI (n = 100), HCs (n = 100) | China | Rv1860, RV3881c, Rv2031c, and Rv3803c | Random forest | 93.3% in training cohort and 95% in validation cohort | 97.7% in training cohort and 80% in validation cohort | 0.981 in training cohort and 0.949 in validation cohort |
Cao et al. [265] | Training cohort, ATB (n = 20), LTBI (n = 20); validation cohort, ATB (n = 92), LTBI (n = 93), HCs (n = 94) | China | Rv1408, R0248, Rv2026c, Rv2716, Rv2031c, Rv2928, and Rv2121c | Logistic regression and hierarchical clustering | 96.77% in training cohort and 93.33% in validation cohort | 93.75% in training cohort and 93.1% in validation cohort | 0.9844 in training cohort and 0.9810 in validation cohort |
Peng et al. [266] | TBI (n = 100), LTBI (n = 60), HCs (n = 44) | China | 15 MTB antigen-specific antibodies | Logistic regression model and hierarchical clustering | 85.4% | 90.3% | 0.944 |
Delemarre et al. [267] | Discovery cohort: ATB (n = 20), LTBI (n = 40), HCs (n = 20); validation cohort: ATB (n = 12 + 31), LTBI (n = 20 + 20) | USA | CCL1, CXCL10, VEGF, and ADA2 | Logistic regression | 95% in discovery cohort, 75% and 100% in validation cohort 1 and 2 | 90% in discovery cohort, 100% and 30% in validation cohort 1 and 2 | Unknown |
Luo et al. [268] | Training cohort: ATB (n = 468), LTBI (n = 424); Test set, ATB (n = 121), LTBI (n = 102); validation cohort: ATB (n = 125), LTBI (n = 138) | China | ESAT-6, CFP-10, IFN-γ, ESR, Hs-CRP | Random forest and bagged ensemble algorithms | 98.85% in Training cohort; 93.39% in Test set; 92.80% in validation cohort | 95.65% in training cohort; 91.18% in Test set; 89.86% in validation cohort | 0.995 in training cohort; 0.978 in Test set; 0.963 in validation cohort |
Morris et al. [269] | Discovery cohort: TB (n = 146), LTBI (n = 146) other diseases (OD) (n = 146); validation cohort: TB (n = 122), OD (n = 127) | Sub-Saharan Africa | Fibrinogen, alpha-2-macroglobulin, CRP, MMP-9, transthyretin, complement factor H, IFN-γ, IP-10, and TNF-α | Random forest and logistic regression | 92% in the test set | 71% in the test set | 0.84 |
Agranoff et al. [270] | Training cohort: ATB (n = 102), HCs (n = 91); validation cohort: ATB (n = 77), HCs (n = 79) | UK | Transthyretin, C-reactive protein, Neopterin, and serum amyloid A | Support vector machine and tree classification | 93.5% | 94.9% | Unknown |
Luo et al. [271] | Discovery cohort: ATB (n = 50), LTBI (n = 49), HC (n = 50); validation cohort: ATB (n = 28), LTBI (n = 24), HCs (n = 26) | China | Eotaxin, MDC, and MCP-1 | No | 87.76% | 91.84% | 0.94 |