wgd—simple command line tools for the analysis of ancient whole-genome duplications

1In this era of whole-genome sequencing, many ancient whole-genome duplication (WGD) events have been uncovered across the eukaryotic tree of life. One of the main approaches for revealing ancient WGDs using genomic data is the construction of whole paranome KS distributions (e.g.;;; ), where KS is the synonymous distance or the estimated number of synonymous substitutions per synonymous site. Under the assumption of neutral evolution at synonymous sites, the synonymous distance between two coding sequences serves as a proxy for the divergence time of two sequences. Under a model of continuous small-scale gene duplication (SSD) and loss of duplicated copies not under selection, a whole paranome KS distribution is expected to show an exponential decay of the number of retained duplicates in function of age. Against this background of SSDs, large-scale duplication events, such as WGDs, are visible as peaks in the number of retained duplicates at a particular age.

2Several issues compromise the use of KS distributions for WGD inference, and these were extensively addressed in. When high-quality genome assemblies are available, gene colinearity (often called synteny) based analyses may further aid in unveiling WGDs or large segmental duplications. WGDs are expected to leave large blocks with high intragenomic colinearity, and paralogs located in such colinear segments (anchor pairs) can therefore be traced back more reliably to a particular event, enabling their use for downstream analyses such as molecular dating or functional analysis.

3While these methods have been used frequently in genomics research, no comprehensive and user-friendly software is available to perform these analyses, and researchers have often resorted to custom pipelines. Here, we fill this gap with an integrated suite for KS and colinearity based analysis of ancient WGDs. We briefly discuss the methods implemented here, but refer to the documentation and Supplementary Material for more information.

4Delineation of paralogous gene families and one-to-one orthologs starts from all-versus-all BLASTp similarity searches or precomputed BLAST results and is performed using ‘wgd mcl’. For whole paranome delineation, MCL is then used to cluster sequences in paralogous gene families. One-to-one orthologs are determined using the commonly employed reciprocal best hit strategy.

5A KS distribution for a set of paralogous families or one-to-one orthologs can be constructed using the ‘wgd ksd’ subcommand, and we closely follow the approach used by. We refrain from a full description of the methodology here and refer to the Supplementary Material instead.

6When high-quality structural genome annotations are available, the ‘wgd syn’ tool allows the identification of intragenomic colinear blocks and their corresponding anchor pairs using I-ADHoRe 3.0. Whole-genome syntenic dotplots are generated, and if a KS distribution is provided, KS-colored dotplots and anchor pair KS distributions are generated (Fig. 1).

7Downstream analyses of KS distributions have often consisted in fitting statistical models and visualizing these. We provide tools (‘wgd kde’) for fitting kernel density estimates (KDEs). Importantly, we apply a correction for boundary effects, which are often neglected but may lead to artificial peaks in low KS regions. As peaks derived from WGDs are expected to be approximately log-normally distributed, Gaussian mixture models (GMMs) have also been used frequently to analyze KS distributions. We provide tools (‘wgd mix’) for fitting mixtures of log-normal components using different inference algorithms, implemented using the scikit-learn python library. Common approaches to determine the optimal number of components are provided, using the Akaike or Bayesian information criterion, however we would like to warn prospective users to carefully interpret ‘significant’ components, as these GMMs may strongly overfit the empirical distribution.

8Lastly, we provide tools for (interactive) visualization of histograms and KDEs in ‘wgd viz’ (Fig. 1). These tools allow visualization of multiple KS distributions for comparative purposes as well as modification of key visualization parameters such as the histogram bin-width or the KDE bandwidth. We encourage researchers to modify and explore the influence of these to guide careful analysis of the distributions and to prevent misinterpretations of KDE or histogram artifacts as biologically interesting features.

9We provide, to our knowledge, the first comprehensive toolshed for KS and colinearity based analysis of WGDs in an easy to use and freely available package named wgd. We hope that, besides being a useful tool for researchers, it will also aid in preventing common pitfalls and misinterpretations when analyzing putative WGDs in genomic data.

10This work was supported by the European Union Seventh Framework Programme (FP7/2007-2013) under European Research Council Advanced Grant Agreement 322739—DOUBLEUP [to Y.V.d.P]; and a PhD Fellowship of the Research Foundation—Flanders (FWO) [to A.Z.].

11Conflict of Interest: none declared.