TY - JOUR
T1 - Diagnostic yield of nine user-friendly bioinformatics tools for predicting Mycobacterium tuberculosis drug resistance: A systematic review and network meta-analysis
AU - Chen, Ya-Li
AU - He, Yu
AU - Dahl, Victor Naestholt
AU - Yu, Kan
AU - Zhang, Yan-An
AU - Guan, Cui-Ping
AU - Wang, Mao-Shui
PY - 2025/4/21
Y1 - 2025/4/21
N2 - To compare the diagnostic yield of various bioinformatics tools for predicting Mycobacterium tuberculosis drug resistance. A systematic review of PubMed, Embase, Scopus, Web of Science, CINAHL and the Cochrane Library was performed to identify studies reporting the effectiveness of bioinformatic tools for predicting resistance to anti-tuberculosis (TB) drugs. Data were collected and pooled using random-effects meta-analysis and Bayesian network meta-analysis (NMA). Summary receiver operating characteristic curves (SROCs) analysis were performed, and superiority index (SI) and area under the curve (AUC) were calculated. Thirty-three studies evaluated 9 different bioinformatics tools for predicting resistance to 14 anti-TB drugs. NMA and SROCs demonstrated that TBProfiler, TGS-TB, Mykrobe, PhyResSE, and SAM-TB all exhibited satisfactory performance. Remarkably, TBProfiler stood out with its exceptional ability to predict resistance to the majority of anti-TB drugs, including isoniazid (SI: 3.39 [95% confidence interval (CI): 0.20, 11.00]; AUC: 0.97 [0.95, 0.98]), rifampicin (SI: 6.38 [0.60, 15.00]; AUC: 0.99 [0.98, 1.00]), ethambutol (SI: 5.15 [0.60, 13.00]; AUC: 0.96 [0.94, 0.97]), streptomycin (SI: 3.67 [0.60, 11.00]; AUC: 0.97 [0.95, 0.98], amikacin (SI: 2.49 [0.14, 11.00]; AUC: 0.97 [0.96, 0.99]), kanamycin (SI: 2.26 [0.14, 9.00]; AUC: 0.98 [0.97, 0.99]), levofloxacin (SI: 1.87 [0.11, 9.00]; AUC: 0.95 [0.93, 0.97]), and prothionamide (SI: 2.73 [0.20, 7.00]; AUC: 0.87 [0.84, 0.90]). Meanwhile, Mykrobe demonstrated superior accuracy specifically for moxifloxacin (SI: 3.96 [0.11, 13.00]; AUC: 0.97 [0.95, 0.98]). Lastly, TGS-TB had the best efficacy in predicting resistance to pyrazinamide (SI: 12.53 [1.67, 17.00]; AUC: 0.97 [0.95, 0.98]), capreomycin (SI: 4.22 [0.08, 15.00]; AUC: 1.00 [0.98, 1.00]), and ethionamide (SI: 2.15 [0.33, 7.00]; AUC: 0.96 [0.94, 0.98]). TBProfiler, TGS-TB, Mykrobe, PhyResSE and SAM-TB have all demonstrated outstanding accuracy in predicting resistance to anti-TB drugs. In particular, TBProfiler stood out for its exceptional performance in predicting resistance to most anti-TB drugs, while TGS-TB excelled in predicting resistance to pyrazinamide and certain second-line drugs. The efficacy of SAM-TB requires further investigation to fully establish its reliability and effectiveness. To ensure the accuracy and reliability of genotypic drug susceptibility testing, bioinformatics tools should be refined and adapted continuously to accommodate novel and current resistance-associated mutations.
AB - To compare the diagnostic yield of various bioinformatics tools for predicting Mycobacterium tuberculosis drug resistance. A systematic review of PubMed, Embase, Scopus, Web of Science, CINAHL and the Cochrane Library was performed to identify studies reporting the effectiveness of bioinformatic tools for predicting resistance to anti-tuberculosis (TB) drugs. Data were collected and pooled using random-effects meta-analysis and Bayesian network meta-analysis (NMA). Summary receiver operating characteristic curves (SROCs) analysis were performed, and superiority index (SI) and area under the curve (AUC) were calculated. Thirty-three studies evaluated 9 different bioinformatics tools for predicting resistance to 14 anti-TB drugs. NMA and SROCs demonstrated that TBProfiler, TGS-TB, Mykrobe, PhyResSE, and SAM-TB all exhibited satisfactory performance. Remarkably, TBProfiler stood out with its exceptional ability to predict resistance to the majority of anti-TB drugs, including isoniazid (SI: 3.39 [95% confidence interval (CI): 0.20, 11.00]; AUC: 0.97 [0.95, 0.98]), rifampicin (SI: 6.38 [0.60, 15.00]; AUC: 0.99 [0.98, 1.00]), ethambutol (SI: 5.15 [0.60, 13.00]; AUC: 0.96 [0.94, 0.97]), streptomycin (SI: 3.67 [0.60, 11.00]; AUC: 0.97 [0.95, 0.98], amikacin (SI: 2.49 [0.14, 11.00]; AUC: 0.97 [0.96, 0.99]), kanamycin (SI: 2.26 [0.14, 9.00]; AUC: 0.98 [0.97, 0.99]), levofloxacin (SI: 1.87 [0.11, 9.00]; AUC: 0.95 [0.93, 0.97]), and prothionamide (SI: 2.73 [0.20, 7.00]; AUC: 0.87 [0.84, 0.90]). Meanwhile, Mykrobe demonstrated superior accuracy specifically for moxifloxacin (SI: 3.96 [0.11, 13.00]; AUC: 0.97 [0.95, 0.98]). Lastly, TGS-TB had the best efficacy in predicting resistance to pyrazinamide (SI: 12.53 [1.67, 17.00]; AUC: 0.97 [0.95, 0.98]), capreomycin (SI: 4.22 [0.08, 15.00]; AUC: 1.00 [0.98, 1.00]), and ethionamide (SI: 2.15 [0.33, 7.00]; AUC: 0.96 [0.94, 0.98]). TBProfiler, TGS-TB, Mykrobe, PhyResSE and SAM-TB have all demonstrated outstanding accuracy in predicting resistance to anti-TB drugs. In particular, TBProfiler stood out for its exceptional performance in predicting resistance to most anti-TB drugs, while TGS-TB excelled in predicting resistance to pyrazinamide and certain second-line drugs. The efficacy of SAM-TB requires further investigation to fully establish its reliability and effectiveness. To ensure the accuracy and reliability of genotypic drug susceptibility testing, bioinformatics tools should be refined and adapted continuously to accommodate novel and current resistance-associated mutations.
U2 - 10.1371/journal.pgph.0004465
DO - 10.1371/journal.pgph.0004465
M3 - Article
C2 - 40258039
SN - 2767-3375
VL - 5
JO - PLoS Global Public Health
JF - PLoS Global Public Health
IS - 4
M1 - e0004465
ER -
