Abstract
Transposable elements (TEs) are major genomic components in most eukaryotic genomes and play an important role in genome evolution. However, despite their relevance the identification of TEs is not an easy task and a number of tools were developed to tackle this problem. To better understand how they perform, we tested several widely used tools for de novo TE detection and compared their performance on both simulated data and well curated genomic sequences. As expected, tools that build TE-models performed better than k-mer counting ones, with RepeatModeler beating competitors in most datasets. However, there is a tendency for most tools to identify TE-regions in a fragmented manner and it is also frequent that small TEs or fragmented TEs are not detected. Consequently, the identification of TEs is still a challenging endeavor and it requires a significant manual curation by an experienced expert. The results will be helpful for identifying common issues associated with TE-annotation and for evaluating how comparable are the results obtained with different tools.