Very few sites can reshape a phylogenetic tree

The history of animal evolution, and the relative placement of extant animal phyla in this history is, in principle, testable from phylogenies derived from molecular sequence data. Though datasets have increased in size and quality in the past years, the contribution of individual genes (and ultimately amino acid sites) to the final phylogeny is unequal across genes. Here we demonstrate that by removing a small fraction of sites that strongly favor one topology, it is possible produce a highly-supported tree of an alternate topology. We explore this approach using a dataset for animal phylogeny, and create a highly-supported tree with a monophyletic group of sponges and ctenophores, a topology not usually recovered. As nearly all gene sets are neither standardized nor representative of the entire genome, we conclude that there are only two possible ways to remedy this problem. One solution would need to use a fixed set of genes, which though not representative, is at least standardized. The other would be to construct phylogenies using all genes, thus limiting analysis to species with sequenced genomes.