Integration of string and de Bruijn graphs for genome assembly.

Abstract

String and de Bruijn graphs are two graph models used by most genome assemblers. At present, none of the existing assemblers clearly outperforms the others across all datasets. We found that although a string graph can make use of entire reads for resolving repeats, de Bruijn graphs can naturally assemble through regions that are error-prone due to sequencing bias. We developed a novel assembler called StriDe that has advantages of both string and de Bruijn graphs. First, the reads are decomposed adaptively only in error-prone regions. Second, each paired-end read is extended into a long read directly using an FM-index. The decomposed and extended reads are used to build an assembly graph. In addition, several essential components of an assembler were designed or improved. The resulting assembler was fully parallelized, tested and compared with state-of-the-art assemblers using benchmark datasets. The results indicate that contiguity of StriDe is comparable with top assemblers on both short-read and long-read datasets, and the assembly accuracy is high in comparison with the others. https://github.com/ythuang0522/StriDe : ythuang@cs.ccu.edu.tw Supplementary data are available at Bioinformatics online.