De novo assembly of ultra-deep sequencing data

Life scientists and bio-informaticians have struggled with insufficient amount of sequencing data since the beginning of Sanger sequencing in the seventies. As a consequence, most of the de novo assembly methods that have been proposed are designed to deal with low coverage sequencing and unbalanced depth of coverage. The situation is now about to change. The cost of sequencing has been decreasing so much that it is interesting to think about the possibility to have "as much sequencing data as we want". When the sequencing will be so cheap that scientists can decide about their desired depth of coverage without being worried about cost, the following question arises: assuming today's sequencing error rate, does higher depth of coverage necessarily lead to a better quality assembly? In this study, we demonstrate for the first time that current state-of-the-art assemblers are unable to handle ultra-deep (i.e., 1,000-10,000x) depth of coverage. We then propose a new method to build high quality assemblies from ultra-deep sequencing data. Our approach is based on "data slicing": we split a large dataset into "slices", then assemble each slice individually using a off-the-shelves assembler. Our tool then merges optimally the individual assemblies. Experimental results show that our method can improve significantly the quality of the assemblies, when compared to the assemblies of the individual slices.