Avoiding Missing Data Biases in Phylogenomic Inference: An Empirical Study in the Landfowl (Aves: Galliformes)

Authors:
Publisher website
Google Scholar
Abstract
Production of massive DNA sequence datasets is transforming phylogenetic inference, but best practices for analyzing such datasets are not well established. One uncertainty is robustness to missing data, particularly in coalescent frameworks. To understand the effects of increasing matrix size and loci at the cost of increasing missing data, we produced a 90 taxon, 2.2 megabase, 4800 locus sequence matrix of landfowl using target capture of ultraconserved elements. We then compared phylogenies estimated with concatenated maximum likelihood, quartet-based methods executed on concatenated matrices, and gene tree reconciliation methods across five thresholds of missing data. Results of maximum likelihood and quartet analyses were similar, well-resolved, and demonstrated increasing support with increasing matrix size and sparseness. Conversely, gene tree reconciliation produced unexpected relationships when we included all informative loci, with certain taxa placed towards the root compared to other approaches. Inspection of these taxa identified a prevalence of short average contigs, which potentially biased gene tree inference and caused erroneous results in gene tree reconciliation. This suggests the more problematic missing data in gene-tree based analyses are partial sequences rather than entire missing sequences from locus alignments. Limiting gene tree reconciliation to the most informative loci solved this problem, producing well-supported topologies congruent with concatenation and quartet methods. Collectively, our analyses provide a well-resolved phylogeny of landfowl, including strong support for previously problematic relationships such as those among junglefowl (Gallus), and clarify the position of two enigmatic galliform genera (Lerwa, Melanoperdix) not sampled in previous molecular phylogenetic studies.

This publication has 112 references indexed in Scilit: