Machine Learning for detection of viral sequences in human metagenomic datasets

BMC Bioinformatics

Table 2 Micro- and macro-average performance measures across 19 experiments using feed-forward neural networks

Method	Class	Precision	Recall	F1-score
Micro-average	Non virus	0.97	1.00	0.99
Micro-average	Virus	0.69	0.13	0.21
Macro-average	Non virus	0.96	1.00	0.98
Macro-average	Virus	0.43	0.12	0.19

The results are from the best model according to area under the ROC curve (ROC_micro=0.790). This model used two 1024 units FC layers with Relu nonlinearity, 0.25 dropout rate and class_weight_power 0.25 (see Additional file 3). All networks were trained for 10 epochs, using Adam optimizer with 10e−4 initial learning rate that was multiplied with 0.95 after each epoch

ISSN: 1471-2105