Additional file 2 of Comparison of 16S and whole genome dog microbiomes using machine learning

Additional file 2: Supplemental Figure S1: A. The increased detection rate in WGS data compared to 16S is maintained at full coverage. B. The detected members of each predominant phylogenetic group at any level. The highest membership across the four predominant phyla in 16S (red) and WGS (blue) was identified in WGS at full coverage. Supplemental Figure S2: The alpha diversity rarefaction of the 16S data set shows the distribution of Fath’s Phylogenetic Diversity (Faith’s PD) index over increasing sequencing depths. The high protein, low carbohydrate diet (HPLC) shows a diminished community richness compared to the baseline diet (BASE) and low protein, high carbohydrate (LPHC) diets, which share similar distributions of phylogenetic diversity at higher sequencing depths. Supplemental Figure S3: A. The beta diversity (Bray-Curtis Dissimilarity) metric multidimensional scaling reveals dissimilarities in the compositional differences of high protein, low carbohydrate (HPLC) and low protein, high carbohydrate (LPHC) WGS microbial communities, and the most dissimilar communities have a higher abundance of B. coprocola. B. A similar analysis of WGS gene family communities indicates HPLC and LPHC diets have greater dissimilarity. These dissimilar communities have a higher LeuB relative abundance, which is coenriched with B. coprocola. Supplemental Figure S4: According to the performance metrics for 12 classification algorithms, the random forest achieves the highest overall percent accuracy in classifying WGS samples. The random forest ties with the gradient boosted regression tree for the highest percent accuracy in classifying 16S samples and achieves the lowest log loss