Long read sequencers portend the possibility of producing reference quality genomes not only because the reads are long, but also because sequencing errors and read sampling are almost perfectly random. However, the error rates are as high as 15%, necessitating an efficient algorithm for finding local alignments between reads at a 30% difference rate, a level that current algorithm designs cannot handle or handle inefficiently. In this paper we present a very efficient yet highly sensitive, threaded filter, based on a novel sort and merge paradigm, that proposes seed points between pairs of reads that are likely to have a significant local alignment passing through them. We also present a linear expected-time heuristic based on the classic O(nd) difference algorithm [1] that finds a local alignment passing through a seed point that is exceedingly sensitive, failing but once every billion base pairs. These two results have been combined into a software program we call DALIGN that realizes the fastest program to date for finding overlaps and local alignments in very noisy long read DNA sequencing data sets and is thus a prelude to de novo long read assembly.
展开▼
