TY - JOUR T1 - Exploring Trade-offs in Scalable Phylogenetic Placement Methods JF - bioRxiv DO - 10.1101/2022.05.23.493012 SP - 2022.05.23.493012 AU - Gillian Chu AU - Tandy Warnow Y1 - 2022/01/01 UR - http://biorxiv.org/content/early/2022/05/25/2022.05.23.493012.abstract N2 - Phylogenetic placement is the problem of placing “query” sequences into an existing tree (called a “backbone tree”) whose leaves are aligned sequences, and has applications to updating large trees and microbiome analysis. While substantial advances have been made in developing methods for phylogenetic placement, to date the most accurate approaches (e.g., pplacer and EPA-ng) are based on maximum likelihood, and these methods tend to have computational challenges when the backbone tree is large. Of the two, EPA-ng can scale to larger backbone tree sizes than pplacer (which seems to be limited to about 5,000-leaf backbone trees), but pplacer seems to have better accuracy than EPA-ng when it can run. Divide-and-conquer methods have been developed to address the limited scalability of pplacer, which operate by finding a small subtree of the backbone tree for the given query sequence, and then placing into that small subtree; SCAMPP is a recent development that shows particular benefits. Another approach, which is specific for pplacer, is taxtastic, which provides numeric model parameters in a form that helps pplacer run on larger datasets. In this study, we examine the potential of using both these approaches for scaling pplacer to large datasets, exploring the impact on accuracy as well as on running time and memory usage. We show that the combination of techniques (i.e., pplacer-taxtastic-SCAMPP) produces the best accuracy of all placement methods to date, with excellent speed and reduced memory usage. Finally, we explore how changing the subtree size associated with the SCAMPP framework changes the runtime-accuracy trade-off, and discuss avenues for future research. Our software for pplacer-taxtastic-SCAMPP is available at https://github.com/gillichu/PLUSplacer-taxtastic.Competing Interest StatementThe authors have declared no competing interest. ER -