Repeat elements are important components of most eukaryotic genomes. Most existing tools for repeat analysis rely either on high quality reference genomes or existing repeat libraries. Thus, it is still challenging to do repeat analysis for species with highly repetitive or complex genomes which often do not have good reference genomes or annotated repeat libraries. Recently we developed a computational method called REPdenovo that constructs consensus repeat sequences directly from short sequence reads, which outperforms an existing tool called RepARK. One major issue with REPdenovo is that it doesn’t perform well for repeats with relatively high divergence rates or low copy numbers. In this paper, we present an improved approach for constructing consensus repeats directly from short reads. Comparing with the original REPdenovo, the improved approach uses more repeat-related k-mers and improves repeat assembly quality using a consensus-based k-mer processing method.

We compare the performance of the new method with REPdenovo and RepARK on Human, Arabidopsis thaliana and Drosophila melanogaster short sequencing data. And the new method fully constructs more repeats in Repbase than the original REPdenovo and RepARK, especially for repeats of higher divergence rates and lower copy number. We also apply our new method on Hummingbird data which doesn’t have a known repeat library, and it constructs many repeat elements that can be validated using PacBio long reads.

We propose an improved method for reconstructing repeat elements directly from short sequence reads. The results show that our new method can assemble more complete repeats than REPdenovo (and also RepARK). Our new approach has been implemented as part of the REPdenovo software package, which is available for download at https://github.com/Reedwarbler/REPdenovo.