MSR-CA -- Efficient De Novo Genome Assembler For Long and Short Read Data

The onset of Second Generation Sequencing technologies that produce high coverage of the genome by short reads at a very low cost has prompted development of new assembly approaches. Several new assembly algorithms based on deBruijn graphs have been recently published. In this presentation we describe a new approach, which has the computational efficiency of the de Bruijn graph and the robustness of the overlap-based assembly strategies. We transform the paired end Illumina reads into a much smaller number of longer Super-reads. The use of super-reads allows us to assemble mixtures of Illumina, 454, and Sanger data, making it one of the few assemblers capable of handling such mixtures. We call this approach the Maryland Super-Reads Celera Assembler (MSR-CA). It is available as open-source code at ftp://ftp.genome.umd.edu/pub/MSR-CA/. We evaluate the performance of the MSR-CA assembler against two most widely used assemblers for Illumina data: Allpaths-LG and SOAPdenovo using two data sets: a bacteria (Rhodobacter sphaeroides) and chromosome 16 of the mouse (Mus musculus) genome. We show that MSR-CA is capable to perform on par with Allpaths-LG and significantly better than SOAPdenovo on these sample data sets, when evaluated against the finished sequence. We then show that MSR-CA can significantly outperform both on the contig sizes and the number of assembly errors if 1x or more long reads (we use Sanger reads) are added to the original Illumina data.