RegCloser

RegCloser: Robust regression-based genome gap closing

RegCloser improves draft genome assemblies by closing gaps using a local assembly strategy that models read coordinates and overlaps through linear regression, addressing genomic repeats and chimeric reads that limit k-mer and overlap-layout-consensus methods.

Key Features:

• Local Assembly via Linear Regression: Represents read coordinates as parameters and read overlaps as observations within a linear regression framework for gap closure.
• Robust Regression Model: Formulates local DNA assembly as a robust parameter estimation problem using a customized procedure that minimizes a convex global Huber loss function to reduce the impact of false overlaps.
• Sparse Linear System Optimization: Iteratively solves a sparse system of linear equations to obtain the global optimum for accurate assembly.
• Tandem Repeat Resolution: Resolves copy number of tandem repeats, increasing assembly completeness and contiguity.
• Long-Read Integration: Applies to long-read data to improve draft genomes and increase contig N50, with potential integration into layout modules of long-read assemblers.

Scientific Applications:

• Genome Assembly Improvement: Enhances completeness and continuity of genome assemblies by resolving complex regions containing repeats and chimeric reads.

Methodology:

RegCloser restricts the search range for optimal overlaps according to insert sizes within a linear regression framework. It performs robust parameter estimation by minimizing a convex global Huber loss and iteratively solving sparse linear systems to mitigate errors from false overlaps.