Genome-wide quantitative analysis of DNA methylation from bisulfite sequencing data

Abstract

Here we present the open-source R/Bioconductor software package BEAT (BS-Seq Epimutation Analysis Toolkit). It implements all bioinformatics steps required for the quantitative high-resolution analysis of DNA methylation patterns from bisulfite sequencing data, including the detection of regional epimutation events, i.e. loss or gain of DNA methylation at CG positions relative to a reference. Using a binomial mixture model, the BEAT package aggregates methylation counts per genomic position, thereby compensating for low coverage, incomplete conversion and sequencing errors.

Availability and implementation: BEAT is freely available as part of Bioconductor at www.bioconductor.org/packages/devel/bioc/html/BEAT.html. The package is distributed under the GNU Lesser General Public License 3.0.

Fig. 1.

(a) Methylation estimates and epimutation calls on a DNA segment. For all regions with sufficient read coverage, the black curves show the methylation estimates for a single cell sample (top), a reference sample (bottom) and their difference (middle). Regions with methylating epimutations are marked in red, while regions with demethylating epimutations are marked in blue. Samples used for our analysis in this article were obtained from neuronal cells of young mice (data unpublished). (b) Scatterplot of methylation estimates of a multi-cell reference sample (x-axis) versus those of a sample (y-axis) for all common regions with sufficient coverage. Each dot represents a single region that is covered by both samples. Red dots indicate methylating epimutations in the sample, while blue dots indicate demethylating epimutations in the sample. Four dots representing exemplary regions with epimutations at the corresponding boundary value ranges for demethylating and methylating epimutations have been annotated with their values of methylated (k) and total (n) counts. Note that there exists no boundary line separating the red and the blue region because our Bayesian model assigns different methylation estimates to tuples (k1, n1), (k2, n2) with equal empirical methylation level k1/n1 = k2/n2