Fig. 2

Importance of compositional bias correction in sparse metagenomic data. a M-A pots of 16S reconstructions (from high sequencing depth, whole metagenome shotgun sequencing experiments) from two technical replicates each from the Tara oceans project [10] generated for the Southern and South Atlantic Oceans. In all subplots, x-axis plots for each feature, its average of the logged proportions in the two compared samples; y-axis plots the corresponding differences. The red dashed line indicates the median log fold change, which is 0 across the technical replicates. b M-A plots of the same replicates but plotted across the two oceans. The median of the log-fold change distribution is clearly shifted. A few dominant taxa in the South Atlantic Ocean (circled) are attributable for driving this overall apparent differences in the observed fold changes. The Tara 16s dataset, reconstructed from very deep whole metagenome shotgun experiments of oceanic samples, albeit boasting of an average 100,000 16S contributing reads per sample, still encourages a median 88% feature absence per sample