RegCloser: a robust regression approach to closing genome gaps

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Closing gaps in draft genomes is crucial for complete and continuous genome assemblies.
• Ubiquitous genomic repeats pose challenges for existing gap-closing methods like de Bruijn graphs and overlap-layout-consensus.
• Chimeric reads can introduce errors in k-mer representations and false overlaps, hindering assembly accuracy.

Purpose of the Study:

• To introduce RegCloser, a novel local assembly approach for robust genome gap closing.
• To address challenges posed by genomic repeats and chimeric reads in genome assembly.
• To improve the accuracy, completeness, and contiguity of genome assemblies.

Main Methods:

• RegCloser employs a linear regression model representing read coordinates and overlaps.
• It searches for optimal overlaps within a restricted range based on insert sizes.
• A customized robust regression procedure using a Huber loss function minimizes the influence of false overlaps.

Main Results:

• RegCloser accurately resolves the copy number of tandem repeats on simulated and real datasets.
• The method achieves superior genome assembly completeness and contiguity compared to other popular tools.
• Application to a plateau zokor draft genome significantly increased contig N50, demonstrating effectiveness with long reads.

Conclusions:

• RegCloser is a competitive and effective tool for genome gap closing.
• The robust regression approach shows potential for integration into long-read assembly layout modules.
• The RegCloser software is publicly available for use in genomic research.