HapBridge: A Methylation-Guided Approach for Correcting Switch Errors and Bridging Phased Blocks in Long-Read Phasing

Hui Chen, Fan Nie, Feng Luo, Jianxin WangbioRxiv 2025 · DOI

TL;DR

This paper introduces HapBridge, a novel computational framework that leverages DNA methylation patterns to improve long-read haplotype phasing. By correcting switch errors and bridging disconnected phased blocks, HapBridge significantly enhances phasing accuracy and contiguity, particularly in challenging genomic regions.

What Problem Does This Paper Address?

Haplotype phasing, the process of assigning genetic variants to their respective parental chromosomes, is crucial for understanding genetic variation, regulatory mechanisms, and disease. While long-read sequencing technologies have advanced this field, current phasing methods still struggle with two major challenges: switch errors (incorrectly flipping the assigned haplotype for a segment of the genome) and gaps in phasing (disconnected "phased blocks") especially in regions with low genetic variation (low heterozygosity). Existing methods often rely solely on sequence variants, which can be sparse. While some approaches have explored complementary data sources like Hi-C or RNA sequencing, these add complexity and cost. DNA methylation, however, exhibits haplotype-specific patterns and is directly measurable from long-read sequencing data, offering a promising, intrinsic source of linkage information. However, previous attempts to integrate methylation into phasing have been limited or sensitive to initial phasing accuracy.

Methods

HapBridge is a two-module framework designed to be integrated into existing SNV-based phasing pipelines.

  1. Switch-Error Correction: This module identifies haplotype-specific methylation sites. It analyzes methylation signals from reads that have already been assigned a haplotype (haplotagged). By clustering methylation values based on these assigned haplotypes and calculating the distance between these clusters, HapBridge can identify sites where methylation patterns are consistently different between the two haplotypes. It then uses a Discrepancy Score to pinpoint SNPs where flipping the assigned haplotype would lead to a more consistent methylation pattern on either side, thereby detecting and correcting switch errors.
  2. Block Bridging: This module connects previously disconnected phased blocks. It first assigns haplotypes to untagged reads (reads not yet assigned to a specific haplotype) using the identified haplotype-specific methylation sites. As more reads are tagged, the methylation information is refined. The core of this module involves identifying high-confidence methylation anchors between blocks. It calculates a consistency score between pairs of methylation sites that co-occur on reads spanning the gap. Sites with high consistency are prioritized, and the process iteratively refines these anchors to bridge the blocks.

Datasets

The study evaluated HapBridge using publicly available datasets from two major long-read sequencing platforms:

  • Oxford Nanopore (ONT) R9/R10: Datasets were used at various coverage levels (80x, 60x, 30x).
  • PacBio HiFi: Datasets were also used at different coverage levels (60x, 30x).

The performance was benchmarked against the Genome in a Bottle (GIAB) SNP benchmark (v4.2.1). The paper also mentions using HG001 and HG002 samples.

Key Insights and Ideas

  • Methylation as an Intrinsic Linkage Signal: The paper strongly advocates for DNA methylation as a robust and inherent signal for phasing, reducing reliance on external data or sparse genetic variants.
  • Complementary Nature of Modules: The switch-error correction and block-bridging modules work synergistically. Correction improves local accuracy within blocks, while bridging enhances global contiguity, demonstrating that both are essential for optimal phasing.
  • Robustness Across Platforms and Coverages: HapBridge shows consistent performance improvements across different sequencing technologies (ONT and PacBio) and varying sequencing depths, highlighting its generalizability.
  • Drop-in Replacement Potential: HapBridge is designed to integrate seamlessly with existing SNV-based phasing tools, making it an accessible upgrade for current pipelines.

Key Results

HapBridge demonstrated significant improvements over existing methods, particularly when integrated with HapCUT2 and WhatsHap:

  • Switch Error Reduction: HapBridge reduced switch errors by 3.07-18.72% compared to MethPhaser. For example, on ONT R9 80x data, HapCUT2-HapBridge achieved 645 switch errors, outperforming HapCUT2 (715) and HapCUT2-MethPhaser (738).
  • N50 Length Improvement: HapBridge improved N50 length by 5.84-68.61% compared to MethPhaser. For instance, on ONT R9 80x data, HapCUT2-HapBridge increased N50 by 142.5% over HapCUT2 and 68.6% over HapCUT2-MethPhaser.
  • Phased Block N50: The phased block N50 metric, which measures the contiguity of correctly phased regions, showed substantial gains with HapBridge, consistently outperforming other methods across datasets.
  • Assembly Performance: When integrated into the HapDup assembly pipeline, HapDup-HapBridge consistently achieved higher phase block N50 values, indicating improved assembly continuity without introducing significant errors.

Limitations and Future Work

  • Coverage Dependence: While robust, the paper notes that switch error rates can be higher for HapCUT2-HapBridge under very low coverage conditions (e.g., 30x). The thresholds for identifying haplotype-specific methylation sites are adjusted based on coverage.
  • Parameter Tuning: The performance of HapBridge relies on user-defined thresholds for identifying haplotype-specific methylation sites. While default values are provided, optimal tuning might be necessary for specific datasets.
  • Future Work: The authors suggest that HapBridge can be further integrated into other long-read phasing and assembly pipelines. The framework's ability to leverage methylation signals could also be explored for other genomic analyses beyond phasing.

Summary updated

Abstract

Abstract Long-read sequencing has substantially advanced haplotype phasing yet continues to face challenges in low-heterozygosity regions and switch errors caused by read noise. DNA methylation exhibits haplotype-specific patterns, providing complementary linkage information, but existing phasing algorithms have not fully leveraged these signals owing to their variability. Here, we present Hap-Bridge , a methylation-guided phasing framework that enhances single-nucleotide variant (SNV)-based phasing by detecting and correcting switch errors and bridging adjacent phased blocks . Evaluations on Oxford Nanopore R9/R10 and PacBio HiFi datasets show that HapBridge reduces switch errors by 3.07–18.72% and improves N50 length by 5.84–68.61% compared with Meth-Phaser, achieving higher phasing accuracy and contiguity. These results demonstrate that integrating methylation with sequence variation can provide a robust and intrinsic linkage signal that effectively improves haplotype continuity in long-read sequencing. HapBridge is publicly available at GitHub https://github.com/Humonex/HapBridge .
BibTeX
@misc{chen2025hapbridge,
  author = {Hui Chen and Fan Nie and Feng Luo and Jianxin Wang},
  title = {{HapBridge: A Methylation-Guided Approach for Correcting Switch Errors and Bridging Phased Blocks in Long-Read Phasing}},
  year = {2025},
  doi = {10.1101/2025.11.07.687303},
  howpublished = {bioRxiv},
}

Loading related content...

Figure