Fig. 3

Percentage of contigs identified as viral in the viral (true positive rate, blue) and microbial (false positive rate, orange) datasets, sorted by the average difference in detection rate between eight viral and microbial paired datasets in seawater (A), soil (B), and gut (C) biomes. Colors of the tool names as in Fig. 1. Convolutional neural network tools outperformed other machine learning tools, while the homology-only tools performed the worst. Receiver operating characteristic (ROC) curves of seven methods based on their virus identification scores in seawater (D), soil (E), and gut biomes (F). The dashed diagonal line represents classification according to random chance. The area under the ROC curve (AUC) of each tool is listed in Additional file 2: Table S9