| Popis: |
Complex repetitive regions (also called segmental duplications) in eukaryotic genomes often contain essential functional and regulatory information. Despite remarkable algorithmic progress in genome assembly in the last twenty years, modern de novo assemblers still struggle to accurately reconstruct these highly repetitive regions. When sequenced reads will be long enough to span all repetitive regions, the problem will be solved trivially. However, even the third generation of sequencing technologies on the market cannot yet produce reads that are sufficiently long (and accurate) to span every repetitive region in large eukaryotic genomes. In this work, we introduce a novel algorithm called RAmbler to resolve complex repetitive regions based on high-quality long reads (i.e., PacBio HiFi). We first identify repetitive regions by mapping the HiFi reads to the draft genome assembly and by detecting unusually high mapping coverage. Then, (i) we compute the k-mers that are expected to occur only once in the genome (i.e., single copy k-mers, which we call unikmers), (ii) we barcode the HiFi reads based on the presence and the location of their unikmers, (iii) we compute an overlap graph solely based on shared barcodes, (iv) we reconstruct the sequence of the repetitive region by traversing the overlap graph. We present an extensive set of experiments comparing the performance of RAmbler against Hifiasm, HiCANU and Verkko on synthetic HiFi reads generated over a wide range of repeat lengths, number of repeats, heterozygosity rates and depth of sequencing (over 140 data sets). Our experimental results indicate that RAmbler outperforms Hifiasm, HiCANU and Verkko on the large majority of the inputs. We also show that RAmbler can resolve several long tandem repeats in Arabidopsis thaliana using real HiFi reads. |
