Computational protocol: Transposable element detection from whole genome sequence data

Similar protocols

Protocol publication

[…] When it comes to detecting mutations, especially somatic mutations, different methods and/or different parametrisations yield markedly different results [–], and transposable element detection is no exception []. Publications presenting new tools often include comparisons where a number of competing methods are run by the authors of the new tool. While valuable, these experiments may not reflect optimal parametrisations of the competing tools for the dataset used as a basis of comparison, whereas by virtue of having developed a novel method, the authors will have better parametrisations of their own tools, leading to the usual outcome of the new tool outperforming previously published methods.To illustrate the extent of the differences in TE insertion calls from different methods run on the same data, we present comparisons between somatic TE detections from three recent studies. In each case, two different methods were used to call mutations on the same data, yielding substantial overlap and an equally if not more substantial amount of non-overlap. Importantly, these calls were generated by the developers of their respective TE detection methods. Coordinates and sample identities were obtained from the supplemental information of the respective studies, and one [] needed to be converted from hg18 to hg19 coordinates via liftOver. Insertion coordinates were padded by +/- 100 bp and compared via BEDTools v2.23. Lee et al. [] (Tea) and Helman et al. [] (TranspoSeq) share 7 samples, Tubio et al. [] (TraFiC) and Helman et al. (TranspoSeq) share 15 samples. No samples are shared between Lee et al. and Helman et al. The overall Jaccard distance between TranspoSeq and Tea results across shared samples was 0.573 (Additional file and Additional file : Table S2a), and between TranspoSeq and TraFiC the distance was 0.741 (Additional file and Additional file : Table S2b), indicating that TranspoSeq and Tea seem to yield more similar results than between TranspoSeq and TraFiC. Summing counts for intersected insertion calls and method-specific calls yields the overlaps shown in Fig. . While this comparison is somewhat cursory and high-level, it is clear there is a substantial amount of difference in the results of these methods: in both comparisons, more insertions are identified by a single program than by both programs. Given that all three studies report a high validation rate (greater than 94 %) where samples were available for validation, this may reflect a difficulty in tuning methods for high sensitivity while maintaining high specificity. This also suggests that perhaps an ensemble approach combining calls across all three (or more) methods may be preferable where high sensitivity is required.Fig. 2In addition to the tools already highlighted, a rapidly increasing number of tools exist with the common goal of detecting transposable element insertions from WGS data. As indicated in Table , these include purpose-built methods aimed specifically at transposable elements in addition to more general methods that identify a wide variety of structural alterations versus a reference genome, transposable element insertions included. Table  is not intended to represent an exhaustive listing of currently existing methods - the OMICtools website (http://omictools.com/) currently supports an up-to-date database of TE detection tools, and the Bergman lab website also hosts a list of transposable element detection tools which includes tools aimed at a wide variety of applications, a subset of which are relevant for TE detection from WGS data []. […]

Pipeline specifications

Software tools liftoveR, BEDTools, TranspoSeq, TraFiC
Application WGS analysis