RegScaf

RegScaf applies regression to scaffold genome assemblies by estimating contig positions and gap distances from linking-read alignments using robust statistical methods.

Key Features:

• Regression-based scaffolding: Parameterizes contig positions on the genome through a linear model that uses observed linking distances as measurements of position differences.
• Kernel density estimation and multimodal clustering: Analyzes the distribution of distances from read alignments with kernel density estimation to identify multiple modes and cluster orientation-supported links by linking distance mode.
• Robust parameter estimation: Estimates model parameters by minimizing a global loss defined as a trimmed sum of squares.
• Least trimmed squares estimator: Employs the least trimmed squares estimate with a high breakdown value to automatically exclude erroneous linking distances.
• Contig boundary effect mitigation: Addresses inaccuracies arising from the contig boundary effect, particularly in repetitive regions, via clustering of linking distances and robust regression.
• Improved gap estimation and repeat resolution: Reduces extreme errors in gap estimates and improves resolution of repeat regions in scaffolds.
• Compatibility with large genomes and TGS reads: Demonstrates adaptability for scaffolding large genomes and third-generation sequencing (TGS) long reads.

Scientific Applications:

• Scaffold accuracy improvement: Improves order, orientation, and gap-distance estimates between contigs in genome assemblies.
• Gap estimation refinement: Reduces extreme errors in estimated gap sizes between contigs.
• Repeat-region resolution: Enhances scaffold correctness in repetitive genomic regions by distinguishing multiple linking-distance modes.
• Benchmarking and validation: Applicable to performance comparison on synthetic and real datasets for scaffolding tools.
• Large-genome and TGS scaffolding: Applied to scaffolding workflows for large genomes and third-generation sequencing long reads.

Methodology:

Compute distance distributions from read alignments using kernel density estimation; identify multiple modes and cluster orientation-supported links; parameterize contig positions with a linear model treating pairwise distances as observations of position differences; estimate parameters by minimizing a trimmed sum of squares via the least trimmed squares estimator to exclude erroneous linking distances.