Skip to main content
Fig. 5 | BMC Bioinformatics

Fig. 5

From: Machine Learning for detection of viral sequences in human metagenomic datasets

Fig. 5

The trade-off between precision and recall for virus vs. non-virus classification when changing the classification threshold. A sample is classified as virus only if p(virus) >threshold. Dotted lines depict the precision and recall for virus class at different thresholds for individual folds (i.e. metagenomics experiments) in LOEO cross validation. Blue line depicts the micro average across the experiments, red line illustrates macro average. The vertical and horizontal dashed lines exemplify what performance can be obtained when boosting precision at the expense of recall. Blue dashed line shows that at 90% accuracy we can obtain 5.67% recall. Cyan dashed line shows that a threshold giving us 95% precision would yield 3.74% recall. The models yielding highest area under ROC curve are used for this graph

Back to article page