Additional File 2: The effects of tree-construction methods on the VAW-UniFrac results

For all the three methods studied in this paper, UniFrac, weighted UniFrac (W-UniFrac) and variance adjusted weighted UniFrac (VAW-UniFrac), a phylogenetic tree of the sequences from the communities is needed. Many different tree construction methods are available including maximum parsimony, distance based methods such as neighbor-joining, and maximum likelihood trees. A natural question is how the different tree-constructing methods affect the subsequent analysis? Lozupone et al.[1] studied the effects of seven tree construction methods on UniFrac and W-UniFrac. They showed that the results from both UniFrac and W-UniFrac are robust to different treeconstructing methods. To investigate the effects of various tree-constructing methods on VAW-UniFrac, we applied four different tree-constructing methods to eight 16S rRNA gene sequence data sets (the Sanger sequencing data sets in Application 3 [2]) and used principal coordinates analysis (PCoA) [3] to show the results. The data sets were described in the main text.