Benchmarking Transposable Element Annotation Methods for Creation of a Streamlined, Comprehensive Pipeline

Abstract

20Sequencing technology and assembly algorithms have matured to the point that high-21 quality de novo assembly is possible for large, repetitive genomes. Current assemblies traverse 22 transposable elements (TEs) and allow for annotation of TEs. There are numerous methods for 23 each class of elements with unknown relative performance metrics. We benchmarked existing 24 programs based on a curated library of rice TEs. Using the most robust programs, we created a 25 comprehensive pipeline called Extensive de-novo TE Annotator (EDTA) that produces a 26 condensed TE library for annotations of structurally intact and fragmented elements. EDTA is 27 open-source and freely available: https://github.com/oushujun/EDTA. 28 Keywords 29 Transposable element; Annotation; Genome; Benchmarking; Pipeline 30 31Long-read sequencing (e.g., PacBio and Oxford Nanopore) and assembly scaffolding 50 (e.g., Hi-C and BioNano) techniques have progressed rapidly within the last few years. These 51 innovations have been critical for high-quality assembly of the repetitive fraction of genomes. In 52 fact, Ou et al. [8] demonstrated that the assembly contiguity of repetitive sequences in recent 53 long-read assemblies is even better than traditional BAC-based reference genomes. With these 54 developments, inexpensive and high-quality assembly of an entire genome is now possible. 55Knowing where features (i.e., genes, TEs, etc.) exist in a genome assembly is important 56 4 information for using these assemblies for biological findings. However, unlike the relatively 57 straightforward and comprehensive pipelines established for gene annotation [9][10][11], current 58 methods for TE annotation can be piecemeal, inaccurate, and are highly specific to classes of 59 transposable elements. 60Transposable elements fall into two major classes. Class I elements, also known as 61 retrotransposons, use an RNA intermediate in their "copy and paste" mechanism of 62 transposition [12]. Class I elements can be further divided into long terminal repeat (LTR) 63 retrotransposons, as well as those that lack LTRs (non-LTRs), which include long interspersed 64 nuclear elements (LINEs), and short interspersed nuclear elements (SINEs). Structural features 65 of these elements can facilitate automated de novo annotation in a genome assembly. For 66 example, LTR elements have a 5-bp target site duplication (TSD), while non-LTRs have either 67 variable length TSDs or lack TSDs entirely, being instead associated with deletion of flanking 68 sequences upon insertion [13]. There are also standard terminal sequences associated with 69 LTR elements (i.e., 5'-TG…C/G/TA-3' for LTR-Copia and 5'-TG…CA-3' for LTR-Gypsy 70 elements), and non-LTRs often have a terminal poly-A tail at the 3' end of the element (see [14] 71 for a complete description of structural features of each superfamily). 72The second major class of TEs, Class II elements, also known as DNA transposons, use 73 a DNA intermediate in their "cut and paste" mechanism of transposition [15]. As with Class I 74...

RepeatModeler2: automated genomic discovery of transposable element families

“…Therefore the number of sequences classified as DNA-TIRs by RepeatModeler2 may be inflated and may rather represent variants or fragments of the same family (see Figure 3). The genome of rice, O. sativa, is known to contain almost equal proportions and numbers of DNA-TIR and LTR elements (Ou et al 2019), and this profile is recovered by our RepeatModeler2 library ( Figure 2). In summary, RepeatModeler2 produces libraries that recapitulate the major TE subclass composition of these three model species.…”

“…RepeatModeler2 uses the LTRharvest (Ellinghaus et al 2008) package for structural LTR detection for both its overall sensitivity and speed , Ou et al 2019. LTRharvest is both a discovery and annotation algorithm that does not attempt to group LTR instances into families.…”

“…We benchmarked RepeatModeler2 on model species that have high-quality reference TE libraries: D. melanogaster, D. rerio, and O. sativa. We used Repbase for the D. melanogaster (release 20181026) and D. rerio (release 14.01) libraries, and the manually-improved library for O. sativa from (Ou et al 2019). We used RepeatMasker and parseRM (https://github.com/4ureliek/Parsing-RepeatMasker-Outputs/blob/master/parseRM.pl) to estimate the percentage of the genome masked by each subclass for the manually-curated and RepeatModeler2 libraries.…”

“…Other TE discovery programs exist, but either focus on finding instances in the genome instead of family consensus sequence building (e.g. Berthelier et al 2018, Ou et al 2019, or are challenging to install and use (Flutre et al 2011). RepeatModeler2 is easy to install and run, as we provide a container version to avoid installing independently all dependencies.…”

Gapless assembly of maize chromosomes using long read technologies

“…The manually curated transposable element library (maizeTE11222019) derived from the Maize TE Consortium (MTEC; https://github.com/oushujun/MTEC) was used as the base TE library. Novel TEs of the maize Ab10 genome not included in the MTEC library were structurally identified using the EDTA pipeline (v1.6.5) 41 with parameters "-species maize -curatedlib maizeTE11222019". The MTEC library augmented with Ab10 specific TEs was used to annotate TE fragments using RepeatMasker.…”