A hierarchical network heuristic for solving the orientation problem in genome assembly

In the past several years, the problem of genome assembly has received considerable attention from both biologists and computer scientists. An important component of current assembly methods is the scaffolding process. This process involves building ordered and oriented linear collections of contigs (continuous overlapping sequence reads) called scaffolds and relies on the use of mate pair data. A mate pair is a set of two reads that are sequenced from the ends of a single fragment of DNA, and therefore have opposite mutual orientations. When two reads of a mate-pair are placed into two different contigs, one can infer the mutual orientation of these contigs. While several orientation algorithms exist as part of assembly programs, all encounter challenges while solving the orientation problem due to errors from mis-assemblies in contigs or errors in read placements. In this paper we present an algorithm based on hierarchical clustering that independently solves the orientation problem and is robust to errors. We show that our algorithm can correctly solve the orientation problem for both faux (generated) assembly data and real assembly data for R. sphaeroides bacteria. We demonstrate that our algorithm is stable to both changes in the initial orientations as well as noise in the data, making it advantageous compared to traditional approaches. Author Summary Constructing an organism’s entire DNA sequence from raw genome sequencing data, like the data produced in the Human Genome Project, is a challenging task. The type of data generated in the sequencing process has changed substantially over the years as a result of various technological improvements. The computer programs that convert such data into assembled sequencing must continuously be revised to keep pace with the changing nature of the data. This paper builds upon current methods from the emerging field of network science to develop a new way of analyzing and correcting sequencing data. We show that our algorithm is both more robust to erroneous data, and more accurate overall, compared to current techniques.