Skip to content

bsseq

Whole-genome bisulfite and long-read methylation analysis in R/Bioconductor — the Hansen-lab workhorse for processing, smoothing, and differentially testing methylation from WGBS, and (as of bsseq >=1.45.1) the substrate for allele-specific methylation analysis from Oxford Nanopore long-read sequencing.

  • Maintainer: Kasper D. Hansen (Johns Hopkins Bloomberg School of Public Health) — kasperdanielhansen@gmail.com
  • Bioconductor: bsseq
  • Source: github.com/hansenlab/bsseq
  • Vignettes: bsseq Analysis
  • Preprint (ASM extension): biorxiv 2025.04.30.651558 — "Sample-specific CpG loci are important for accurate long-read methylation analysis"
  • License: Artistic-2.0
  • Status at GBCC2025: methods talk on the allele-specific-methylation (ASM) extension from ONT long reads

Talk at GBCC2025

  • Speaker: Kasper D. Hansen (Johns Hopkins)
  • Day / session: Day 2 (Tue Jun 24, 2025) — Oral Session 2B, Grace Auditorium (chair: Vincent Carey)
  • Talk title: "Analyzing allele-specific methylation using Oxford Nanopore long-read DNA sequencing"
  • Slides: TBD (Zenodo deposit typically 2-4 weeks post-conference)
  • Video: GBCC2025 playlist
  • Abstract / schedule: GBCC2025 program

What it does

bsseq stores and analyzes whole-genome cytosine methylation data — originally from WGBS but now extended to Oxford Nanopore long reads. Core capabilities: smoothing methylation estimates (BSmooth), F-statistic-based differentially methylated region (DMR) calling, and methylation summarization across regions. The Hansen-lab 2025 extension (bsseq >=1.45.1, slated for Bioconductor 3.22) adds allele-specific methylation (ASM) analysis from ONT long reads, leveraging the fact that long reads physically link CpG sites to the haplotype they sit on — letting an analyst test for methylation differences between maternal and paternal alleles without needing trio data or a separate phasing step.

How it works

Core idea. BSmooth fits a local-likelihood (locally weighted) regression of methylation along the genome, borrowing strength across nearby CpGs within a CpG cluster so sparse-coverage loci still get a stable estimate. A two-group t-statistic with optional large-scale-bias correction is then thresholded, and contiguous super-threshold runs are emitted as DMRs.

Inputs / outputs. Input is a BSseq object holding per-CpG coverage matrices and methylated-read (M) counts from WGBS aligners (the reference implementation uses Merman output); for the ONT extension, methylation calls from long-read pipelines are loaded into the same container. Output is smoothed methylation estimates per CpG and a DMR table with areaStat, meanDiff, width, and cluster membership.

Key innovation. Adaptive local smoothing that preserves spatial resolution while sharing power across neighboring CpGs, with explicit handling of large-scale methylation shifts (common in cancer). The 2025 extension layers haplotype-resolved ASM on top by exploiting the physical phasing native to ONT long reads — no trio or separate phasing step required.

Parameters worth knowing. - h — half-width of the smoothing window (default 1000 bp). - ns — minimum CpGs per window (default 70). - maxGap — distance at which CpG clusters are split (default 100 Mb). - mc.cores — parallelization across samples.

Canonical example. The vignette uses BS.cancer.ex (chr21–22 cancer vs. normal colon, ABI SOLiD 50 bp): 

library(bsseq)
data(BS.cancer.ex)
BS.fit   <- BSmooth(BS.cancer.ex, mc.cores=6)
BS.tstat <- BSmooth.tstat(BS.fit, group1=c("C1","C2","C3"), group2=c("N1","N2","N3"))
dmrs     <- dmrFinder(BS.tstat, cutoff=c(-4.6, 4.6))

Where it fits in the corpus

  • AACR 2026: axis = bioinfo / AI methods; ASM is one of the more interesting non-cancer-specific extensions of long-read methylation analysis with cancer-imprinting and tumor-suppressor-silencing implications
  • EuroBioC 2025: no direct counterpart; methylation-from-long-reads is GBCC-side this year
  • ISMB 2026: scaffold; expect overlap on long-read methylation methods
  • Nextflow Summit 2026: not mentioned

Notes

bsseq is a mature, foundational Bioconductor package (in continuous development since ~2012). The ASM extension is the news at GBCC2025 — it folds ONT-specific phasing-aware logic into the same BSseq container that the WGBS community has been using for over a decade, which keeps the analysis ergonomics consistent for users who are switching modalities.