An assembly and alignment-free method of phylogeny reconstruction from next-generation sequencing data

Abstract

BackgroundNext-generation sequencing technologies are rapidly generating whole-genome datasets for an increasing number of organisms. However, phylogenetic reconstruction of genomic data remains difficult because de novo assembly for non-model genomes and multi-genome alignment are challenging.ResultsTo greatly simplify the analysis, we present an Assembly and Alignment-Free (AAF) method (https://sourceforge.net/projects/aaf-phylogeny) that constructs phylogenies directly from unassembled genome sequence data, bypassing both genome assembly and alignment. Using mathematical calculations, models of sequence evolution, and simulated sequencing of published genomes, we address both evolutionary and sampling issues caused by direct reconstruction, including homoplasy, sequencing errors, and incomplete sequencing coverage. From these results, we calculate the statistical properties of the pairwise distances between genomes, allowing us to optimize parameter selection and perform bootstrapping. As a test case with real data, we successfully reconstructed the phylogeny of 12 mammals using raw sequencing reads. We also applied AAF to 21 tropical tree genome datasets with low coverage to demonstrate its effectiveness on non-model organisms.ConclusionOur AAF method opens up phylogenomics for species without an appropriate reference genome or high sequence coverage, and rapidly creates a phylogenetic framework for further analysis of genome structure and diversity among non-model organisms.Electronic supplementary materialThe online version of this article (doi:10.1186/s12864-015-1647-5) contains supplementary material, which is available to authorized users.

Phylogenomics of 42 tomato chloroplasts using assembly and alignment-free method

“…On the other hand, identical k-mers could be derived from physically, functionally, or evolutionary different regions of the genome and are therefore not homologous (k-mer homoplasy). Longer k-mers are less likely to suffer from k-mer homoplasy 6 . Thus, the selection of k-mer length is a trade-off between the problem of sensitivity (which requires a smaller k) and k-mer homoplasy (which requires a larger k).…”

“…Some studies have proposed to filter out all k-mers which frequencies are below a given threshold θ . For example by removing k-mers that present less than three copies (θ = 3) can reduce the impact of the sequencing errors 6 . However, as sequencing coverage decreases, a larger fraction of real k-mers will be singletons in the dataset, and therefore filtering will remove real k-mers.…”

“…Filtering out singletons can correct the sequencing error effect with low coverage (between 5-8X), according to genome size 6 . In Figure 2 (A), 5X coverage Illumina sequencing simulation of the 12 cp genomes and with a k-mer length of 25, two peaks may be observed.…”

“…A k-mer is a substring of nucleotides A, C, T and G of length k. As the number of k-mers counted depends on the sequencing coverage and the distribution of the reads on the genome, this frequency table is converted to a table of presence/absence of k-mers among taxa. Then, the phylogenetic distance D between two species is estimated using the metric 6 :…”

“…Such techniques are based on any number of statistical, computational, and biological principles. Recently, Fan et al 6 have developed an assembly and alignment-free (AAF) method for phylogeny reconstruction. This method firstly calculates pairwise genetic distances between two samples of short sequence reads.…”