Additional file 1 of Machine learning-based prediction of antimicrobial resistance and identification of AMR-related SNPs in Mycobacterium tuberculosis

Supplementary Material 1: Figure S1. Proportion of drug-resistant and susceptible isolates of Mycobacterium tuberculosis in the datasets used in this study. The blue group represents drug-susceptible isolates, and the red group represents drug-resistant isolates. Figure S2-S18. Assessment of the performance of the machine learning algorithms in predicting resistance to 17 antibiotics by MTB in 6-fold cross validation settings and interpretability for the ML model. (A) The preformance metrics. (i) training precision, (ii) training recall, (iii) training F1, (iv) test precision, (v) test recall, (vi) 10-fold CV (cross validation), (vii) Loo CV (leave-one-out cross validation), (viii) au ROC (area under ROC curve) and ix) au PR (area under precision recall curve). ‘All’ denotes all SNPs for training (as in the cross-validation partitioning), ‘Intersection’ refers to AMR SNPs that consistently ranked high across all 6 rounds of cross-validation, and ‘Random’ refers to randomly sampled SNPs. (B) The interpretability for the ML models. (i) The importance of each SNP in building the final predictive model. (ii) SHAP summary plot of SNPs contributing to the GBC model. X-axis shows the average of the absolute SHAP values. (iii) SHAP values of SNPs in the GBC model. Y-axis lists the different SNPs, and X-axis shows the SHAP values. Each dot in the plot represents an MTB isolate. The color of the dots from blue to red shows the feature values from low to high. (iv) SHAP force plot for explaining of a single MTB isolate’s prediction result. SHAP, SHapley Additive exPlanations; GBC, Gradient Boosting Classifier. Figure S19. The potential common genes for MTB resistance to different antibiotics. The ribbons connecting drugs to mutation genes represent the associations. The longer the length of the outer ring, the more antibiotics (genes) are associated with the gene (antibiotic). Figure S20. Proportion of drug-resistant and susceptible isolates of MTB in two independant datasets. (A) MTB isolates collected from Chennai, India (NCBI Bioproject accession ID: PRJNA741102) and MTB isolates from Israel (NCBI Bioproject accession ID: PRJNA957554). Note: R, resistant. S, susceptible. I, intermidiate. NA, not available. (B) Proportions of Different Lineages in Two Independent MTB Datasets.