Maximum likelihood inference of reticulate evolutionary histories

Abstract

Hybridization plays an important role in the evolution of certain groups of organisms, adaptation to their environments, and diversification of their genomes. The evolutionary histories of such groups are reticulate, and methods for reconstructing them are still in their infancy and have limited applicability. We present a maximum likelihood method for inferring reticulate evolutionary histories while accounting simultaneously for incomplete lineage sorting. Additionally, we propose methods for assessing confidence in the amount of reticulation and the topology of the inferred evolutionary history. Our method obtains accurate estimates of reticulate evolutionary histories on simulated datasets. Furthermore, our method provides support for a hypothesis of a reticulate evolutionary history inferred from a set of house mouse (Mus musculus) genomes. As evidence of hybridization in eukaryotic groups accumulates, it is essential to have methods that infer reticulate evolutionary histories. The work we present here allows for such inference and provides a significant step toward putting phylogenetic networks on par with phylogenetic trees as a model of capturing evolutionary relationships. reticulate evolution | incomplete lineage sorting | phylogenetic networks | maximum likelihood P hylogenetic trees have long been a mainstay of biology, providing an interpretive model of the evolution of molecules and characters and a backdrop against which comparative genomics and phenomics are conducted. Nevertheless, some evolutionary events, most notably horizontal gene transfer in prokaryotes and hybridization in eukaryotes, necessitate going beyond trees (1). These events result in reticulate evolutionary histories, which are best modeled by phylogenetic networks (2). The topology of a phylogenetic network is given by a rooted, directed, acyclic graph (rDAG) that is leaf-labeled by a set of taxa ( Fig. 1; more details are provided in Model and SI Appendix). Reticulation events result in genomic regions with local genealogies that are incongruent with the speciation pattern. Several methods and heuristics use this incongruence as a signal for inferring reticulation events and reconstructing phylogenetic networks from local genealogies. These methods, which are surveyed elsewhere (2-4), assume that reticulation events are the sole cause of all incongruence among the gene trees and seek phylogenetic networks to explain all of the incongruence. A serious limitation of these methods is that they would grossly overestimate the amount of reticulation in a dataset when other causes of incongruence are at play. Indeed, several recent studies (5-9) have shown that detecting hybridization in practice can be complicated by the presence of incomplete lineage sorting (ILS) (Fig. 1).Recently, a set of methods was devised to analyze data where reticulation and ILS might both be simultaneously at play (10-15). However, these methods are all applicable to simple scenarios of species evolution and mostly assume a known hypothesis about the topol...

Spaces of phylogenetic networks from generalized nearest-neighbor interchange operations

“…Ding et al, 2011). Note that some non-local operations on rooted networks are defined in (Yu et al, 2014) that could be of interest in this context.…”

“…In Strimmer and Moulton (2000) it was proposed to develop methods to build networks by searching through networks so as to optimize likelihood, and some progress has been made in this direction for this and related criteria (see e.g. Jin et al, 2006Jin et al, , 2007Radice, 2011), especially in the recent work by Yu et al (2014). However, although a similar strategy has been successfully employed for several years to search tree space and construct phylogenetic trees (see e.g.…”

Fast and accurate statistical inference of phylogenetic networks using large-scale genomic sequence data

“…Each of the n rows in the alignment A is a sample representing taxon x ∈ X, and each taxon is represented by one or more samples. Similar to other approaches [12,14], we also require an input parameter C r which specifies the number of reticulation nodes in the output phylogeny.…”

“…The scalability of the state of the art falls well short of that required by current phylogenetic studies, where many dozens or hundreds of divergent genomic sequences are common [3]. The most accurate phylogenetic network inference methods performed statistical inference under phylogenomic models [12,14,16] that extended the multi-species coalescent model [17,18]. MPL and SNaQ were among the fastest of these methods while MLE and MLElength were the most accurate.…”

Reconstructing Phylogenetic Level-1 Networks from Nondense Binet and Trinet Sets

“…We expect that our algorithm could be useful in practice as there are programs which can be used to compute trinets from biological data [14,25] (and also binets as subnets of the computed trinets). Some of these programs use optimisation criteria such as likelihood which can be very computationally expensive for large datasets, but which are much more practical for small datasets.…”