Reader Chengxi Ye mentioned to us a few months back that he developed a very efficient genome assembler for PacBio data, which did PacBio assembly of human genome in 6 hours on a desktop computer compared to >400K CPU hours using a cluster. For those, who do not know Chengxi Ye, he was the author of SparseAssembler and also developed ultrafast basecaller BlindCall. The algorithm of SparseAssembler is more relevant in the current context, because he leveraged its compression scheme in the new PacBio/hybrid assembler.
The paper is posted in arxiv and the programs can be downloaded from here.
The genome assembly computational problem is preventing the transition from the prevalent second generation to third generation sequencing technology. The problem emerged because the erroneous long reads made the assembly pipeline prohibitive expensive in terms of computational time and memory space consumption.
In this paper, we propose and demonstrate a novel algorithm that allows efficient assembly of long erroneous reads of mammalian size genomes on a desktop PC. Our algorithm converts the de novo genome assembly problem from the de Bruijn graph to the overlap layout consensus framework. We only need to focus on the overlaps composed of reads that are non-contained within any contigs built with de Bruijn graph algorithm, rather than on all the overlaps in the genome data sets. For each read spanning through several contigs, we compress the regions that lie inside each de Bruijn graph contigs, which greatly lowers the length of the reads and therefore the complexity of the assembly problem. The new algorithm transforms previously prohibitive tasks such as pair-wise alignment into jobs that can be completed within small amount of time. A compressed overlap graph that preserves all necessary information is constructed with the compressed reads to enable the final-stage assembly.
We implement the new algorithm in a proof-of-concept software package DBG2OLC. Experiments with the sequencing data from the third generation technologies show that our method is able to assemble large genomes much more efficiently than existing methods. On a large PacBio human genome dataset we calculated the pair-wise alignment of 54x erroneous long reads of human genome in 6 hours on a desktop computer compared to the 405,000 CPU hours using a clusters, previously reported by Pacific Biosciences. The final assembly results were in comparably high quality.