A performance study of the impact of recombination on species tree analysis

Abstract

BackgroundThe most widely used state-of-the-art methods for reconstructing species phylogenies from genomic sequence data assume that sampled loci are identically and independently distributed. In principle, free recombination between loci and a lack of intra-locus recombination are necessary to satisfy this assumption. Few studies have quantified the practical impact of recombination on species tree inference methods, and even fewer have used genomic sequence data for this purpose. One prominent exception is the 2012 study of Lanier and Knowles. A main finding from the study was that species tree inference methods are relatively robust to intra-locus recombination, assuming free recombination between loci. The latter assumption means that the open question regarding the impact of recombination on species tree analysis is not fully resolved.ResultsThe goal of this study is to further investigate this open question. Using simulations based upon the multi-species coalescent-with-recombination model as well as empirical datasets, we compared common pipeline-based techniques for inferring species phylogenies. The simulation conditions included a range of dataset sizes and several choices for recombination rate which was either uniform across loci or incorporated recombination hotspots. We found that pipelines which explicitly utilize inferred recombination breakpoints to delineate recombination-free intervals result in greater accuracy compared to widely used alternatives that preprocess sequences based upon linkage disequilibrium decay. Furthermore, the use of a relatively simple approach for recombination breakpoint inference does not degrade the accuracy of downstream species tree inference compared to more accurate alternatives.ConclusionsOur findings clarify the impact of recombination upon current phylogenomic pipelines for species tree inference. Pipeline-based approaches which utilize inferred recombination breakpoints to densely sample loci across genomic sequences can tolerate intra-locus recombination and violations of the assumption of free recombination between loci.Electronic supplementary materialThe online version of this article (doi:10.1186/s12864-016-3104-5) contains supplementary material, which is available to authorized users.

Phylogenomic analyses reveal a deep history of hybridization and polyploidy in the Neotropical genus Lachemilla (Rosaceae)

“…Some studies (e.g. Gatesy & Springer, ; Springer & Gatesy, ) argue that recombination might affect coalescent‐based phylogenetic analyses, but simulation studies have shown that methods for species tree inference may be largely robust to intra‐locus recombination (Lanier & Knowles, ; Wang & Liu, ), and a recent empirical study showed that, despite a large amount of recombinant loci (~ 42%), ASTRAL‐II still recovered the same topology with these loci included or excluded from species tree analyses (Folk et al ., ). With respect to our ASTRAL‐II analyses, the only difference with and without recombinant loci is again in the placement of the Orbiculate clade.…”

Build a Better Bootstrap and the RAWR Shall Beat a Random Path to Your Door: Phylogenetic Support Estimation Revisited

“…Examples include statistical inference of species trees under non-i.i.d. models of sequence evolution (Thorne et al, 1992;Wang and Liu, 2016), machine learning using temporal models (Breiman, 1996), and many, many more -far too many to list here. Table 5.…”

Delimiting Coalescence Genes (C-Genes) in Phylogenomic Data Sets

“…Also, recombination breakpoints and ILS that are associated with increasingly divergent taxa will exhibit higher levels of homoplasy. Still, coalescence analyses will be compromised if the critical issue of c-gene boundaries is simply ignored [ 21 ]. One strategy may be to search for recombination breakpoints with small subsets of taxa that are associated with individual coalescent events (e.g., human-chimp-gorilla or three species of Equus ) and then to combine these together to produce a c-gene map for the full set of taxa.…”

“…In another relevant simulation study, Wang and Liu [ 21 ] compared data-driven methods for detecting recombination and reported that pipelines that use direct analysis of the data to explicitly infer breakpoints and delineate c-genes (‘recombination-free intervals’) result in greater species tree accuracy with the summary coalescence method ASTRAL II relative to pipelines that pre-process loci into intervals of a given length based on independent estimates of linkage disequilibrium decay. Wang and Liu’s [ 21 ] explicit recombination approach was limited to detecting recombination breakpoints with the four-gamete test (FGT) [ 59 ]. The FGT assumes the infinite sites model, so all violations of the four-gametes rule are interpreted as recombination.…”

“…In Section 5 , we ask whether complete protein-coding sequences are appropriate for coalescence analyses. Section 6 reviews two simulation studies [ 20 , 21 ] that ask whether recombination is a problem for summary coalescence methods. We conclude that recombination must matter at some scale if coalescence methods effectively address the problem of ILS.…”