HMD-ARG: hierarchical multi-task deep learning for annotating antibiotic resistance genes

Microbiome

Table 2 The ARG antibiotic classes’ classification results between different methods

	Accuracy	Precision^a	Recall	F1-score
HMD-ARG	0.935	0.950	0.847	0.893
CARD	0.418	0.983	0.452	0.585
DeepARG	0.887	0.914	0.757	0.814
AMRPlusPlus^b	0.675	0.574	0.278	0.283
Meta-MARC^c	0.909	0.750	0.782	0.745

^aThe precision, recall, and F1-score are the macro average over the ARG antibiotic classes
^bAMR++ requires the input in a paired fastq format. So, we simulated the fastq data from our test protein dataset. Details can be seen in the supplementary
^cThe drug class label system of AMR++ and Meta-MARC are different from ours, so we manually converted their labels into ours. Meta-MARC can work with both assembly and raw data; we tested it with the assembled sequences (our test dataset)

ISSN: 2049-2618