Scalable genome scaffolding using integer linear programming

Abstract

The rapidly diminishing cost of genome sequencing is driving renewed interest in large scale genome sequencing programs such as Genome 10K (G10K). Despite renewed interest the assembly of large genomes from short reads is still an extremely resource intensive process. This work presents a scalable algorithms to create scaffolds, or ordered and oriented sets of assembled contigs, which is one part of a practical assembly. This is accomplished using integer linear programming (ILP). In order to process large mammalian genomes we employ non-serial dynamic programming (NSDP) and a hierarchical strategy. Both existing and novel quantitative metrics are used to compare scaffolding tools and gain deeper insight into the challenges of scaffolding. The code is available at: https://bitbucket.org/jrl03001/silp

SILP3: Maximum likelihood approach to scaffolding

“…Note that in the ordering O the estimation of gaps between adjacent contigs uniquely defines the gap g O (i, j) between any pair of connected contigs i and j. SILP2 [3] extracts the maximum subgraph-ordering out of G . Instead, SILP3 is aimed to find an ordering with the maximum support of G -edges.…”

“…The flow of SILP2 [3] consists of the following steps: 1) mapping reads onto contigs 2) scaffolding graph construction 3) maximum likelihood contig orientation via ILP 4) decomposition into paths of orientation compatible edges via bipartite matching 5) maximum likelihood gap estimation II. MAXIMUM LIKELIHOOD ORIENTATION Let G = (V, E) be the scaffolding graph connecting vertices-contigs with edges-read pairs.…”

“…MAXIMUM LIKELIHOOD ORIENTATION Let G = (V, E) be the scaffolding graph connecting vertices-contigs with edges-read pairs. Maximum likelihood contig orientation can be formulated as the following ILP [3]. Let S i be a boolean variable with value being set to 0 if the orientation of contig i remains unchanged.…”

OPERA-LG: Efficient and exact scaffolding of large, repeat-rich eukaryotic genomes with performance guarantees

“…Based on the scaffold assembly formulation from Huson et al [21], this study proposes extensions to an exact algorithm [13] that make it feasible for scaffolding large, repeat-rich genomes in a time and memory-efficient manner [22,23]. The new approach, termed OPERA-LG, was extensively evaluated against state-of-the-art scaffolders (SSPACE [14], SOPRA [16] and BESST [24]) and assembly pipelines (SOAPdenovo [25] and ALLPATHS-LG [26]) on simulated and real datasets.…”

“…In addition to being the first, scalable scaffolding and assembly algorithm 1 with proven performance guarantees (other recent works on scaffolding are either reported to be not scalable or do not have formal guarantees (Dayarian et al 2010;Salmela et al 2011;Lindsay et al 2012)), OPERA-LG incorporates several additional features specifically tailored for producing high quality draft assemblies for large, repeat-rich genomes. These include the ability to simultaneously use data from multiple libraries (a first for stand-alone scaffolding programs and originally implemented in the Celera Assembler at the mate-pair level (Myers et al 2000)) as is typically needed in large assembly projects, an improved edge-length estimation algorithm and an exact extension for scaffolding the repetitive sequences that typically confound assembly tools.…”