First author [Ref.] | ML application | Algorithm | Statistical performance | Strengths | Limitations |
---|---|---|---|---|---|
Nguyen, M. (2018) [17] | Used ML-based NGS-based pAST method to predict the MICs for Klebsiella pneumoniae in order to determine the degree of resistance status on 20 antibiotics | Gradient-boosting tree | Average accuracy: 92.0% | No prior knowledge of the gene content was needed; the continuous phenotype (MIC) was predicted | No external validation |
Khaledi, A. (2020) [18] | Used ML-based NGS-based pAST method to predict the antibiotic susceptibility of Pseudomonas aeruginosa to 4 antibiotics (ceftazidime, meropenem, ciprofloxacin and tobramycin) | Support vector machine | -Using single-nucleotide polymorphism’s information or gene expression data alone: Sensitivity and Predictive value: 0.8–0.9 -In combination: Sensitivity and Predictive value: >0.9 | No prior knowledge of molecular mechanisms of the resistance was needed | No external validation |
Aun, E. (2018) [19] | Used ML-based NGS-based pAST method to predict the antibiotic susceptibility of Pseudomonas aeruginosa to ciprofloxacin | Logistic regression for binary phenotypes (susceptible or resistant); Linear regression for the continuous phenotype (MIC value) | -Logistic regression: Accuracy: 88.0%. F1-measure: 0.88. Sensitivity: 0.90. Specificity: 0.87 -Linear regression: Coefficient of determination (R2): 0.42. Pearson correlation coefficient: 0.68. Spearman correlation coefficient: 0.84 | A simple software ‘PhenotypeSeeker’ was developed; the continuous phenotype (MIC) was predicted | No external validation |