Testing Phylogenetic Placement Accuracy of DNA Barcode Sequences on a Fish Backbone Tree: Implications of Backbone Tree Completeness and Species Representation

Abstract

ABSTRACT

Advancements in DNA sequencing technology have facilitated the generation of a vast number of DNA sequences, posing opportunities and challenges for constructing large phylogenetic trees. DNA barcode sequences, particularly COI, represent extensive orthologous sequences suitable for phylogenetic analysis. Phylogenetic placement analysis offers a promising method to integrate COI data into tree‐building efforts, yet the impacts of backbone tree completeness and species composition remain under‐explored. Using a dataset comprising 27 genes and 4520 species of bony fishes, we assessed the accuracy of phylogenetic inference by “placing” COI sequences onto backbone trees. The backbone tree completeness was varied by subsampling 20%, 40%, 60%, 80%, and 99% of the total species separately, followed by placement of those missing species based on their COI sequences using software packages EPA‐ng and APPLES. We also compared the effects of biased, random, and stratified sampling strategies; the latter ensured the representation of all major lineages (Family) of bony fish. Our findings indicate that the placement accuracy is consistently high across all levels of backbone tree completeness, where 70%–78% missing species are correctly placed (by EPA‐ng) in the same locations as the reference tree derived from the complete data. High completeness produces slightly high placement accuracy, although in many cases the differences are nonsignificant. For example, at the 99% completeness level with stratified sampling, EPA‐ng placed 78% missing species correctly, and when only considering placement with high confidence (LWR > 0.9), the percentage is 87%. Additionally, stratified sampling outperforms random sampling in most cases, and biased sampling has the worst performance. The likelihood‐based EPA‐ng consistently provide higher accurate placements than the distance‐based APPLES. In conclusion, COI‐based placement analysis represents a potential route of using the available vast barcoding data for building large phylogenetic trees.