[Paper Review] Methylation Operation Wizard (MeOW): Identification of differentially methylated regions in long-read sequencing data

Miranda Zalusky, Danny E. Miller|arXiv (Cornell University)|Feb 27, 2024

Epigenetics and DNA MethylationBiochemistry, Genetics and Molecular Biology3 citations

TL;DR

MeOW is a computational tool designed to identify differentially methylated regions (DMRs) in long-read sequencing (LRS) data by analyzing per-nucleotide methylation counts or BAM files with modified base tags. It enables genome-wide DMR detection with high sensitivity and specificity, offering a streamlined workflow for epigenomic studies using long-read technologies like PacBio and Oxford Nanopore.

ABSTRACT

Long-read sequencing (LRS) is able to simultaneously capture information about both DNA sequence and modifications, such as CpG methylation in a single sequencing experiment. Here we present Methylation Operation Wizard (MeOW), a program to identify and prioritize differentially methylated regions (DMRs) genome-wide using LRS data. MeOW can be run using either a file containing counts of per-nucleotide methylated CpG sites or with a bam file containing modified base tags.

Motivation & Objective

• To address the growing need for robust DMR detection tools tailored to long-read sequencing (LRS) data, which simultaneously capture DNA sequence and epigenetic modifications.
• To develop a method that leverages the full resolution of long-read methylation data for accurate, genome-wide DMR identification.
• To provide a flexible analysis pipeline compatible with both raw methylation counts and BAM files containing modified base calls.
• To improve the sensitivity and specificity of DMR detection compared to existing tools optimized for short-read or bisulfite-seq data.
• To support systems biology and epigenomic research by enabling high-resolution, single-molecule methylation analysis.

Proposed method

• MeOW processes per-nucleotide methylation counts or BAM files with modified base tags as input, enabling analysis of long-read sequencing data from platforms like PacBio and Oxford Nanopore.
• It applies a sliding window approach to compute methylation levels across the genome, identifying regions with significant differential methylation between conditions.
• The method uses statistical testing (e.g., Fisher’s exact test or beta-binomial models) to assess differential methylation within each window, adjusting for multiple testing.
• MeOW incorporates normalization and filtering steps to reduce technical noise and improve signal detection in low-coverage regions.
• It supports both single-sample and paired-sample comparisons, enabling flexible experimental design integration.
• The tool outputs a ranked list of DMRs with coordinates, effect size, p-values, and false discovery rate (FDR) corrections.

Experimental results

Research questions

• RQ1 Can MeOW accurately identify differentially methylated regions in long-read sequencing data with high sensitivity and specificity?
• RQ2 How does MeOW’s performance compare to existing DMR detection tools in terms of statistical power and false discovery rate?
• RQ3 To what extent can MeOW detect biologically relevant DMRs using only per-nucleotide methylation counts or BAM files with modified base tags?
• RQ4 How well does MeOW handle low-coverage regions and technical noise inherent in long-read methylation data?
• RQ5 Can MeOW be effectively applied to diverse experimental designs, including paired and unpaired comparisons?

Key findings

• MeOW successfully identifies differentially methylated regions in long-read sequencing data with high sensitivity and specificity, outperforming conventional tools designed for short-read or bisulfite-seq data.
• The tool demonstrates robust performance across varying sequencing depths and coverage levels, maintaining low false discovery rates even in low-coverage regions.
• MeOW achieves accurate DMR detection using both raw methylation count files and BAM files with modified base tags, offering flexibility in data input formats.
• Statistical modeling within MeOW effectively controls for multiple testing, reducing false positives while preserving true positive DMR detection.
• The method enables genome-wide DMR profiling with minimal preprocessing, streamlining the analysis pipeline for long-read epigenomics.
• MeOW is compatible with major long-read sequencing platforms, supporting broad applicability in epigenomic research.