Fig. 3: MD-CNN and SD-CNN model generalize well on held-out test data. | Nature Communications

Fig. 3: MD-CNN and SD-CNN model generalize well on held-out test data.

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

Fig. 3

Performance of CNN models trained on the entire training dataset evaluated on either the entire training dataset or the entire hold-out test dataset (N = 12,848 isolates). a data are presented as mean AUCs ±  95% confidence intervals for first-line (4 drugs) and second-line drugs (7 drugs). b AUC for each drug evaluated on either the entire training or entire hold-out test dataset. Ciprofloxacin and ethionamide (both second-line drugs) were not assessed due to low numbers of resistant isolates.

Back to article page