Fig. 2

Classifier performance on cross-validated sequence datasets. Classification accuracy of 16S rRNA gene V4 subdomain (first row), V1–3 subdomain (second row), full-length 16S rRNA gene (third tow), and fungal ITS sequences (fourth row). a Average F-measure for each taxonomy classification method (averaged across all configurations and all cross-validated sequence datasets) from class to species level. Error bars = 95% confidence intervals. b Average F-measure for each optimized classifier (averaged across all cross-validated sequence datasets) at species level. Violins with different lower-case letters have significantly different means (paired t-test false detection rate-corrected P < 0.05). c correlation between F-measure performance for each method/configuration classification of V4 subdomain (x axis), V1–3 subdomain (y axis), and full-length 16S rRNA gene sequences (z axis). Inset lists the Pearson R² value for each pairwise correlation; each correlation is significant (P < 0.001)