DNA 5-methylcytosine detection and methylation phasing using PacBio circular consensus sequencing

Peng Ni, Fan Nie, Zeyu Zhong, Jinrui Xu, Neng Huang, Jun Zhang, Haochen Zhao, You Zou, Yuanfeng Huang, Jinchen Li, Chuan‐Le Xiao, Feng Luo, Jianxin Wang

Abstract: Long single-molecular sequencing technologies, such as PacBio circular consensus sequencing (CCS) and nanopore sequencing, are advantageous in detecting DNA 5-methylcytosine in CpGs (5mCpGs), especially in repetitive genomic regions. However, existing methods for detecting 5mCpGs using PacBio CCS are less accurate and robust. Here, we present ccsmeth, a deep-learning method to detect DNA 5mCpGs using CCS reads. We sequence polymerase-chain-reaction treated and M.SssI-methyltransferase treated DNA of one human sample using PacBio CCS for training ccsmeth. Using long (≥10 Kb) CCS reads, ccsmeth achieves 0.90 accuracy and 0.97 Area Under the Curve on 5mCpG detection at single-molecule resolution. At the genome-wide site level, ccsmeth achieves >0.90 correlations with bisulfite sequencing and nanopore sequencing using only 10× reads. Furthermore, we develop a Nextflow pipeline, ccsmethphase, to detect haplotype-aware methylation using CCS reads, and then sequence a Chinese family trio to validate it. ccsmeth and ccsmethphase can be robust and accurate tools for detecting DNA 5-methylcytosines.

Introduction

DNA 5-methylcytosine (5mC) is a vital epigenetic mark involved in development and disease. For decades, the "gold standard" for detection has been bisulfite sequencing (BS-seq). However, bisulfite treatment is chemically harsh, damaging DNA and leading to biased genomic coverage. While long-read technologies like Nanopore and PacBio can detect modifications in native DNA by analyzing signal changes, PacBio’s Circular Consensus Sequencing (CCS)—also known as HiFi sequencing—has historically struggled with 5mC detection accuracy. Existing models were often "brittle" or required extremely high coverage to be reliable. The authors developed ccsmeth to unlock the full potential of PacBio HiFi reads for both site-specific methylation and haplotype-aware phasing.

Methods

The ccsmeth framework utilizes kinetic signals generated during PacBio sequencing, specifically Interpulse Duration (IPD) and Pulse Width (PW). The model architecture employs a Bidirectional Gated Recurrent Unit (BiGRU) combined with an Attention mechanism to process these signals.

  1. Read-level prediction: The model extracts a 21-mer sequence context around a CpG site and processes kinetic features from both the forward and reverse strands. The Attention layer focuses on the most informative signals.
  2. Site-level prediction: To estimate methylation frequency at a specific genomic location, ccsmeth aggregates read-level probabilities from multiple overlapping reads, using a secondary BiGRU to incorporate information from neighboring CpG sites.

Additionally, the ccsmethphase pipeline integrates variant calling and phasing to assign methylation states to specific maternal or paternal alleles.

Overview of the ccsmeth workflow, including read generation and model architecture.

Key Insights and Datasets

The study utilized a comprehensive suite of datasets, including PCR-amplified (unmethylated) and M.SssI-treated (methylated) DNA for training, alongside human benchmarks (HG002, CHM13) and a Chinese family trio for validation.

  • Kinetics as "Speed Bumps": The core innovation is treating the DNA polymerase as a sensor; it slows down differently when encountering a methylated base. Deep learning picks up these subtle, non-linear kinetic patterns.
  • Symmetry and Efficiency: Because CpG methylation is typically symmetric, ccsmeth gains power by analyzing both strands of the same molecule. It achieves high correlation (>0.90) with BS-seq using only 10× coverage.
  • Repetitive Regions: Long CCS reads allow ccsmeth to profile methylation in genomic "dark matter" (centromeres and telomeres) where short-read methods typically fail.

Key Results

At the single-molecule level, ccsmeth achieved 0.90 accuracy and an AUC of 0.97. Benchmarking against existing tools like primrose and the HK model showed ccsmeth consistently outperformed them in sensitivity and specificity across various datasets. Furthermore, ccsmethphase successfully identified Allele-Specific Methylation (ASM) in known imprinted regions, showing high consistency with both BS-seq and Nanopore data.

Benchmarking ccsmeth against existing models (HK model and primrose) across multiple datasets.

Limitations and Future Work

The current model assumes symmetric CpG methylation, meaning it cannot yet detect hemimethylation or non-CpG methylation (CpH). Additionally, it is not yet designed to distinguish between 5mC and other modifications like 5hmC. Performance also remains dependent on subread depth; lower depth increases kinetic noise. As sequencing chemistry improves (e.g., PacBio Revio), these deep-learning approaches will likely become even more powerful for integrated genome and epigenome assembly.

Summary updated

Abstract

Long single-molecular sequencing technologies, such as PacBio circular consensus sequencing (CCS) and nanopore sequencing, are advantageous in detecting DNA 5-methylcytosine in CpGs (5mCpGs), especially in repetitive genomic regions. However, existing methods for detecting 5mCpGs using PacBio CCS are less accurate and robust. Here, we present ccsmeth, a deep-learning method to detect DNA 5mCpGs using CCS reads. We sequence polymerase-chain-reaction treated and M.SssI-methyltransferase treated DNA of one human sample using PacBio CCS for training ccsmeth. Using long (≥10 Kb) CCS reads, ccsmeth achieves 0.90 accuracy and 0.97 Area Under the Curve on 5mCpG detection at single-molecule resolution. At the genome-wide site level, ccsmeth achieves >0.90 correlations with bisulfite sequencing and nanopore sequencing using only 10× reads. Furthermore, we develop a Nextflow pipeline, ccsmethphase, to detect haplotype-aware methylation using CCS reads, and then sequence a Chinese family trio to validate it. ccsmeth and ccsmethphase can be robust and accurate tools for detecting DNA 5-methylcytosines.
BibTeX
@inproceedings{ni2023dna,
  author = {Peng Ni and Fan Nie and Zeyu Zhong and Jinrui Xu and Neng Huang and Jun Zhang and Haochen Zhao and You Zou and Yuanfeng Huang and Jinchen Li and Chuan‐Le Xiao and Feng Luo and Jianxin Wang},
  title = {{DNA 5-methylcytosine detection and methylation phasing using PacBio circular consensus sequencing}},
  booktitle = {Nature Communications},
  year = {2023},
  doi = {10.1038/s41467-023-39784-9},
}
Figure