Using Gene Trees with Lineage-Specific Duplicates for Phylogenetic Inference Mitigates the Effects of Long-Branch Attraction

Traditionally, the inference of species trees has relied on orthologs, or genes related through speciation events, to the exclusion of paralogs, or genes related through duplication events. This has led to a focus on using only gene families with a single gene-copy per species, as these families are likely to be composed of orthologs. However, recent work has demonstrated that phylogenetic inference using paralogs is both accurate and allows researchers to take advantage of more data. Here, we investigate a case in which using larger gene families actually increases accuracy compared to using single-copy genes alone. Long-branch attraction is a phenomenon in which taxa with long branches may be incorrectly inferred as sister taxa due to homoplasy. The most common solution to long-branch attraction is to increase taxon sampling to break up long branches. Sampling additional taxa is not always feasible, possibly due to extinction or limited access to high-quality DNA. We propose the use of larger gene families with additional gene copies to break up long branches. Using simulations, we demonstrate that using larger gene families mitigates the impacts of long-branch attraction across large regions of parameter space, especially when either maximum parsimony or maximum likelihood with a misspecified substitution model is used for inference. We also analyze data from Chelicerates, with a focus on assessing support for a sister relationship between scorpions and pseudoscorpions. Previous work has suggested that the failure to recover this relationship is due to long-branch attraction between pseudoscorpions and other lineages. Using data from larger gene families increases support for a clade uniting scorpions and pseudoscorpions, although other resolutions of this relationship continue to have higher support. Together, this work highlights the potential utility of larger gene families in phylogenetic inference.