Quartet-based methods to reconstruct phylogenetic networks

Abstract

Background: Phylogenetic networks are employed to visualize evolutionary relationships among a group of nucleotide sequences, genes or species when reticulate events like hybridization, recombination, reassortant and horizontal gene transfer are believed to be involved. In comparison to traditional distance-based methods, quartet-based methods consider more information in the reconstruction process and thus have the potential to be more accurate.

Results: We introduce QuartetSuite, which includes a set of new quartet-based methods, namely QuartetS, QuartetA, and QuartetM, to reconstruct phylogenetic networks from nucleotide sequences. We tested their performances and compared them with other popular methods on two simulated nucleotide sequence data sets: one generated from a tree topology and the other from a complicated evolutionary history containing three reticulate events. We further validated these methods to two real data sets: a bacterial data set consisting of seven concatenated genes of 36 bacterial species and an influenza data set related to recently emerging H7N9 low pathogenic avian influenza viruses in China.

Conclusion: QuartetS, QuartetA, and QuartetM have the potential to accurately reconstruct evolutionary scenarios from simple branching trees to complicated networks containing many reticulate events. These methods could provide insights into the understanding of complicated biological evolutionary processes such as bacterial taxonomy and reassortant of influenza viruses.

Figure 1

A phylogenetic tree with 6 leaves. Each labeled node indicates a taxon and r is the common ancestor. Branch length indicates evolutionary distance between two taxa.

Figure 2

A phylogenetic network with 3 reticulates. Each node labeled in lower case letter indicates a taxon and r is the common ancestor. There are three reticulate events labeled by A, B, and C. For example, the sequence of taxa h is resulted from concatenating partial sequence from j and partial sequence from k in reticulate event A.

Figure 3

Comparison of four phylogenetic reconstruction methods in tree (A) and network (B) reconstruction. The red bar indicates sensitivity and the other one indicates specificity; the horizontal line indicates the median performances.

Figure 4

QuartetS networks on 11 GC-rich bacteria. Each node indicates a bacteria and the labels in bold font indicate the genus of a group of bacteria.

Figure 5

QuartetS networks on 11 GC-rich and GC-poor bacteria. Each node indicates a bacteria and the labels in bold font indicate the genus of a group of bacteria. The two groups GC-Rich and GC-Poor bacteria are shaded separately.

Figure 6

QuartetS networks on all 36 bacteria. Each node indicates a bacteria and the labels in bold font indicate the genus of a group of bacteria. The two groups GC-Rich and GC-Poor bacteria are shaded separately.

Figure 7

Reassortant network related to H7N9 influenza A viruses.

QuartetA

Figure 8

The influence of thresholdc on sensitivity and specificity of QuartetS for two simulated data. The sensitivity and specificity plotted for all 100 runs with the horizontal line in the bar indicating the median performances. The upper whisker extends from the hinge to the highest value that is within 1.5×IQR of the hinge, where IQR is the inter-quartile range, or distance between the first and third quartiles. The lower whisker extends from the hinge to the lowest value within 1.5×IQR of the hinge. Data beyond the end of the whiskers are outliers and plotted as solid squares.