Repeats are computationally difficult to detect and annotate ab initio because of their abundance, varied features/sequence signatures, many length variants (truncated versions) and clade specificity. Many computational tools have been developed to detect TEs, and the most commonly used approaches can be divided into three categories:
- Library-based methods (e.g. RepeatMasker [15]), that use sequence alignment to search a genome for homologs of known repeats from a database such as Repbase [16], Repbase is a manually curated repeat library of species-specific and pan-species TEs, and cannot be used to identify segmental duplications.
- Signature-based methods, that rely on the fact that each class of TE has a set of unique sequence features such as target site duplications, a poly-A tail, terminal inverted repeats, etc… These methods search for the sequence signatures of the repeat class of interest (e.g. LTR_STRUC [17]). However, because repeat types are so varied, this method is usually only able to identify specific types of TE.
- Ab initio consensus methods, four examples here are RepeatModeler (http://www.repeatmasker.org/RepeatModeler/), REPET [18], Red [19] and PILER [20]. RepeatModeler (RMD) is a de novo package that has been widely used for repeat identification and modeling that combines different programs: RepeatMasker, RepeatScout [21], RECON [22] and TRF (Tandem Repeat Finder) [23]. RepeatMasker identifies and masks interspersed repeats using curated libraries of consensus sequences supported by Dfam; Dfam contains entries corresponding to all Repbase TE entries, and each Dfam entry is represented by a profile hidden Markov model. RECON evaluates pair-wise similarities to build repeat consensus sequences. RepeatScout identifies and uses highly over-represented k-mers as seeds that are extended to produce multiple sequence alignments. However, RMD doesn’t identify the individual sequences used to derive the consensus sequences; making it impossible to confirm or assess the accuracy of the consensus sequences, or to directly analyse the repeat instances in the genome they are derived from.