Quartet Fiduccia–Mattheyses revisited for larger phylogenetic studies

Abstract

Motivation With the recent breakthroughs in sequencing technology, phylogeny estimation at a larger scale has become a huge opportunity. For accurate estimation of large-scale phylogeny, substantial endeavour is being devoted in introducing new algorithms or upgrading current approaches. In this work, we endeavour to improve the QFM (Quartet Fiduccia and Mattheyses) algorithm to resolve phylogenetic trees of better quality with better running time. QFM was already being appreciated by researchers for its good tree quality, but fell short in larger phylogenomic studies due to its excessively slow running time. Results We have re-designed QFM so that it can amalgamate millions of quartets over thousands of taxa into a species tree with a great level of accuracy within a short amount of time. Named QFM Fast and Improved (QFM-FI), our version is 20,000x faster than the previous version and 400x faster than the widely used variant of QFM implemented in PAUP* on larger datasets. We have also provided a theoretical analysis of the running time and memory requirements of QFM-FI. We have conducted a comparative study of QFM-FI with other state-of-the-art phylogeny reconstruction methods, such as QFM, QMC, wQMC, wQFM and ASTRAL, on simulated as well as real biological datasets. Our results show that QFM-FI improves on the running time and tree quality of QFM and produces trees that are comparable with state-of-the-art methods. Availability and Implementation QFM-FI is open source and available at https://github.com/sharmin-mim/qfm_java. Supplementary information Supplementary Information is available at Bioinformatics online.

wQFM-TREE: highly accurate and scalable quartet-based species tree inference from gene trees

“…The effectiveness of such a move, which we call "Gain", is quantified by the differences in bipartition scores before and after a taxon transfer. Mim et al [2023] showed that given Θ(n 4 ) quartets, one iteration of this technique of improving an initial bipartition takes O(n 4 ) time which is very high. Producing an initial bipartition at every bipartitioning step is also a highly time-consuming process.…”

“…In the presence of gene tree heterogeneity, standard methods for estimating species trees, such as concatenation (which concatenates multiple sequence alignments of different genes into a single super-alignment and then estimates a tree from this alignment) can be statistically inconsistent [Degnan et al, 2009, Roch andSteel, 2015], and produce incorrect trees with high support [Kubatko and Degnan, 2007]. Therefore, "summary methods", which operate by computing gene trees from different loci and then combining the inferred gene trees into a species tree, are becoming increasingly popular, and many of them are provably statistically consistent [Avni et al, 2015, Bayzid and Warnow, 2012, Chifman and Kubatko, 2014, Islam et al, 2020, Mahbub et al, 2021, Mim et al, 2023, Mirarab et al, 2014b, Reaz et al, 2014, Snir and Rao, 2010, Zhang, 2011.…”

“…Han and Molloy have recently developed TREE-QMC, a substantially faster version of wQMC, eliminating the need for explicitly computing the weighted quartet distribution displayed by the input gene trees [Han and Molloy, 2023]. Mim et al [2023] have recently presented a fast implementation of the original Quartet Fiduccia-Mattheyses (QFM) algorithm [Reaz et al, 2014], which solves the unweighted version of the maximum quartet consistency problem. In this work, we introduce a novel method, wQFM-TREE (wQFM applied directly to gene trees), employing an innovative technique to conduct the same heuristic search as wQFM, which obviates the need to decompose the input gene trees into their induced quartets.…”

wQFM-DISCO: DISCO-enabled wQFM improves phylogenomic analyses despite the presence of paralogs

“…Many summary methods have been proposed to target ILS and some of them have been shown to be statistically consistent under the multi-species coalescent (MSC) model (Avni et al, 2015; Bayzid and Warnow, 2012; Chifman and Kubatko, 2014; Islam et al, 2020; Mahbub et al, 2021; Mim et al, 2023; Mirarab et al, 2014; Reaz et al, 2014; Snir and Rao, 2010; Zhang, 2011). Other statistically consistent species-tree estimation methods include BEST (Liu, 2008) and *BEAST (Heled and Drummond, 2010), which co-estimate gene trees and species trees from input sequence alignments.…”

“…However, the most scalable and popular approach to date remains a two-step process, where gene trees are first inferred independently from sequence data, and then combined using summary methods that attempt to estimate species trees by summarizing the input gene trees under a model of gene tree discordance. Many summary methods have been proposed to target ILS and some of them have been shown to be statistically consistent under the multi-species coalescent (MSC) model (Avni et al, 2015;Bayzid and Warnow, 2012;Chifman and Kubatko, 2014;Islam et al, 2020;Mahbub et al, 2021;Mim et al, 2023;Snir and Rao, 2010;Zhang, 2011). Other statistically consistent species-tree estimation methods include BEST (Liu, 2008) and *BEAST (Heled and Drummond, 2010), which co-estimate gene trees and species trees from input sequence alignments.…”