Gaps and complex structurally variant loci in phased genome assemblies.

Area of Science:

• Genomics
• Bioinformatics
• Human Genetics

Background:

• Phased human genome assembly has advanced using long-read data and parental or linked-read information.
• Despite progress, current methods like trio-based assembly (e.g., trio-hifiasm) still result in numerous gaps (over 140 per assembly).

Purpose of the Study:

• To analyze assembly gaps, breaks, and misorientations in a diverse set of human haploid assemblies.
• To compare phasing accuracy using Strand-seq versus parental data.
• To identify common locations and types of assembly gaps and assess their impact on protein-coding genes.

Main Methods:

• Analysis of 182 haploid assemblies from 77 human samples.
• Comparison of chromosome-wide phasing accuracy between trio-based methods and Strand-seq.
• Detailed examination of gap locations, focusing on repetitive elements like segmental duplications and satellite DNA.
• Estimation of DNA misorientations and identification of large-scale alignment discontinuities (deletions/insertions).

Main Results:

• Assembly gaps predominantly cluster in large, identical repeats (35.4% segmental duplications, 22.3% satellite DNA, 27.4% GA/AT-rich regions).
• 1513 protein-coding genes overlap assembly gaps, with 231 recurrently affected.
• 6-7 Mbp of DNA are misoriented per haplotype, with 81% corresponding to large inversion polymorphisms.
• Significant large-scale deletions (11.9 Mbp) and insertions (161.4 Mbp) per haploid genome identified, primarily in satellite DNA, but also in 230 euchromatic regions impacting 197 protein-coding genes.

Conclusions:

• Trio-based methods are the gold standard, but Strand-seq offers comparable chromosome-wide phasing accuracy.
• Repetitive regions, particularly segmental duplications and satellite DNA, are major sources of assembly gaps and variations.
• Incompletely assembled and variable regions, including those affecting protein-coding genes, are critical targets for improving genome assembly algorithms and pangenome representations.