NCM 2021: Genome-wide detection of cytosine methylations in plants from nanopore data

Feng presented the tool DeepSignal for methylation calling, focusing on its application to calling 5mC in plant genomes. In comparison to mammalian genomes, in which 5mC is typically found in a CpG context, in plants, 5mC occurs at CpG, CHG, and CHH sites (where H is any base except G). Feng explained that there is a lack of accurate analysis methods for profiling 5mCs in all three contexts. As methylation at CHG/CHH sites is typically at a lower level than at CpG sites, producing high-confidence training data is a particular challenge. Considering these issues, the team developed the DeepSignal-plant algorithm, paying particular attention to how they trained the model. The output of the pipeline is a .BED (bedMethyl format) file containing methylation frequencies of called sites. Feng displayed results, including the demonstration that compared to bisulphite sequencing, DeepSignal-plant detected 1.1% more 5mCs in the Arabidopsis thaliana genome, and 5.3% more 5mCs in the Oryza sativa genome — due to greater coverage in repetitive regions, such as centromeres and telomeres.