Flexible metagenome analysis using the MGX framework

Microbiome

Table 1 Taxonomic classification performance on genus level for benchmark datasets

	Kraken	Kaiju	Centrifuge	MetaPhlAn 2	MGX
RefSeq
True positive	12,059,412	9,329,288	12,611,380	414,943	12,566,362
False positive	18,748	185,899	53,092	7,171	20,698
False negative	1,281,840	3,844,813	695,528	12,937,886	772,940
Sensitivity	0.9039	0.7082	0.9477	0.0311	0.9421
Precision	0.9984	0.9805	0.9958	0.9830	0.9984
Accuracy	0.9027	0.6983	0.9440	0.0311	0.9406
F1 score	0.9488	0.8224	0.9712	0.0602	0.9694
GenBank
True positive	1,851,436	2,592,655	2,175,122	92,383	3,976,270
False positive	398,899	1,230,445	864,989	10,378	734,389
False negative	9,629,665	8,56,900	8,839,889	11,777,239	7,169,341
Sensitivity	0.1613	0.2435	0.1975	0.0078	0.3568
Precision	0.8227	0.6782	0.7155	0.8990	0.8441
Accuracy	0.1558	0.2182	0.1831	0.0078	0.3347
F1 score	0.2697	0.3583	0.3095	0.0154	0.5015

All tools achieve high precision on the RefSeq-derived metagenome, as the source organisms are already included in the relevant classification databases. For the GenBank-based metagenome containing only species not present in the tools’ databases, MetaPhlAn 2 offers high precision but only a very low sensitivity (0.78%), followed by the MGX-provided default pipeline, which ranks highest in sensitivity and accuracy as well as F1 score. Numbers in italics denote best results

ISSN: 2049-2618