Fig. 2: MD-CNN performs comparably to state-of-art WDNN for both first- and second-line drugs. | Nature Communications

Fig. 2: MD-CNN performs comparably to state-of-art WDNN for both first- and second-line drugs.

From: A convolutional neural network highlights mutations relevant to antimicrobial resistance in Mycobacterium tuberculosis

Fig. 2

Results of five-fold cross validation on the training dataset (N = 10,201 isolates) for the four models: WDNN, logistic regression + L2 benchmark, SD-CNN, and MD-CNN. a Data are presented as pooled mean AUCs ± 95% confidence intervals for first-line (5 AUC values per drug, 4 drugs) and second-line drugs (5 AUC values per drug, 9 drugs). b data are presented as individual AUC values for each cross-validation split. The WDNN was not initially trained on levofloxacin or ethionamide and thus was not evaluated for these drugs.

Back to article page