Tischler, G. | Search bioRxiv

bioRxiv2017

Haplotype And Repeat Separation In Long Reads

Resolving the correct structure and succession of highly similar sequence stretches is one of the main open problems in genome assembly. For non haploid genomes this includes determining the sequences of the different haplotypes. For all but the smallest genomes it also involves separating different repeat instances. In this paper we discuss methods for resolving such problems in third generation long reads by classifying alignments between long reads according to whether they represent true or false read overlaps. The main problem in this context is the high error rate found in such reads, which greatly exceeds the amount of difference between the similar regions we want to separate. Our methods can separate read classes stemming from regions with as little as 1% difference.

bioinformatics↗

bioRxiv2017

Non Hybrid Long Read Consensus Using Local De Bruijn Graph Assembly

While second generation sequencing led to a vast increase in sequenced data, the shorter reads which came with it made assembly a much harder task and for some regions impossible with only short read data. This changed again with the advent of third generation long read sequencers. The length of the long reads allows a much better resolution of repetitive regions, their high error rate however is a major challenge. Using the data successfully requires to remove most of the sequencing errors. The first hybrid correction methods used low noise second generation data to correct third generation data, but this approach has issues when it is unclear where to place the short reads due to repeats and also because second generation sequencers fail to sequence some regions which third generation sequencers work on. Later non hybrid methods appeared. We present a new method for non hybrid long read error correction based on De Bruijn graph assembly of short windows of long reads with subsequent combination of these correct windows to corrected long reads. Our experiments show that this method yields a better correction than other state of the art non hybrid correction approaches.

bioinformatics↗