Skip to main content
Premium Trial:

Request an Annual Quote

Researchers Devise New Single-Cell DNA Methylation Prediction Method

NEW YORK (GenomeWeb) – A UK team has developed a computational method for filling in gaps in available DNA methylation profiles produced from individual cells.

The researchers used a so-called deep neural network strategy to develop their method, called DeepCpG, as they reported yesterday in Genome Biology. From there, they applied the approach to single-cell DNA methylation data for a handful of mouse or human cell types, demonstrating that DeepCpG compared favorably with existing strategies for predicting DNA methylation and profiling DNA motifs related to epigenetic variability in the cells.

"Across all cell types, DeepCpG yielded substantially more accurate predictions of methylation states than previous approaches," senior author Oliver Stegle, a group leader at the European Bioinformatics Institute's European Molecular Biology Lab, and his co-authors wrote. DeepCpG "uncovered both previously known and de novo sequence motifs that are associated with methylation changes and methylation variability between cells," they added.

The DeepCpG method brings together DNA modules marked by sequence clues with bidirectional gated recurrent networks that track the presence or absence of methylation at CpG sites in the genome, the team explained. The resulting "joint module" is designed to learn the DNA-CpG interactions as a means of producing a model to predict DNA methylation more generally.

Once trained appropriately, the researchers noted, DeepCpG can be used for everything from methylation prediction in shallow-sequenced cells to analyses aimed at identifying DNA motifs that typically coincide with a given methylation state or with methylation variability between cells.

When they applied DeepCpG to whole-genome, single-cell bisulfite sequence data for 18 mouse embryonic stem cells, for example, the authors found that the approach "yielded more accurate predictions than any of the alternative methods, both genome-wide and in different genomic contexts."

The team subsequently applied DeepCpG to dozens of mouse or human cells that had undergone single-cell bisulfite sequencing or reduced-representation single-cell bisulfite sequencing. They looked at its methylation prediction performance and also its ability to link specific DNA sequence motifs to cytosine methylation features or to gauge the methylation consequences of DNA mutations.

"Several of the motifs discovered by DeepCpG could be matched to known motifs that are implicated in the regulation of DNA methylation," the authors wrote. "The specific motifs that can be discovered are intrinsically limited to motifs that account for variations in a given dataset and hence depend on the considered cell type and latent factors that drive methylation variability."