String Graph Assembly for Diploid Genomes with Long Reads

We use PacBio^® data from whole-genome shotgun sequencing of multiple strains of Arabidopsis thaliana to evaluate the feasibility of using a string graph to assemble diploid heterozygous genomes.  We first demonstrate the generation of high-quality assemblies (N50 ~ 4 Mbp) for inbred strains with a string graph assembly method.  Then, we perform a few computational experiments to understand the properties of the string graph for polymorphic or heterozygous diploid genomes.  Distinct diploid structures in the string graph constructed with long reads are observed and a simple layout algorithm is developed to resolve homologous copies from genomic repeats. This allows constructing long primary contigs from a diploid heterozygous sample.  The structural variations between the haplotypes can be captured by constructing “associated contigs,” which represent alternative paths other than those of the primary contigs in the string graph.  A full set of structure variations (SVs) can be then categorized. These SVs and the primary contigs can be used as the foundation for the downstream process of phasing SVs and SNPs together for full reconstruction of the haplotype sequences.  We successfully applied this strategy to generate heterozygous diploid assemblies (N50 > 2 Mbp) and evaluate the correctness from a data set of mixed inbred Arabidopsis sequencing data.  Finally, we apply this approach to assemble several diploid genomes and propose a general bioinformatics workflow for diploid assembly with long sequence reads.