Interpretation of microbiota-based diagnostics by explaining individual classifier decisions

BMC Bioinformatics

Table 1 Most relevant bacterial species for classification of the gut microbiota sample shown in Fig. 1

Species name	IS-pro peak(s)	Presence	Relevance	Common habitat
Alistipes putredinis	235	+	0.43	Gut
Alistipes finegoldii	396	+	0.4	Gut
	231	+	0.33	Gut
Lachnospiraceae sp.	605	+	0.25	Gut
Bacteroides fragilis	537	+	0.22	Gut
Staphylococcus epidermidis	313	–	0.2	Skin
Odoribacter splanchnicus	307	+	0.18	Gut
Bacteroides sp.	549	+	0.16	Gut
Prevotella sp.	438	+	0.15	Gut
Lachnospiraceae sp.	491	+	0.15	Gut
Alistipes finegoldii	230	–	−0.14	Gut
Bacteroides vulgatus	478	+	0.14	Gut
Streptococcus mitis	296	–	0.13	Skin
Sutterella wadsworthensis	661	+	0.12	Gut
unclassified Bacteroidete	455	+	0.09	Gut
unclassified Proteobacterium	932	+	0.09	–
Escherichia coli	735/828	+	0.08	Gut
unclassified Firmicute	558	+	0.07	Gut

’+’ and’-’ indicate whether a peak is present or absent in a sample, and needs to be coupled with the relevance value for interpretation. Ranking is based on absolute relevance values; Classification was done by a linear SVM

ISSN: 1471-2105