Feng Luo from Clemson University presented at the Nanopore Community Meeting 2021 about “Genome-wide detection of cytosine methylations in plants from nanopore data using deep learning.” They are faculty in computer science at Clemson. They spoke about detecting methylation from nanopore signals. They developed the DeepSignal tool. Nanopore has developed Megalodon, and there is also the nanopolish tool for methylation detection. Mammalian cells can have 5mC mainly in CpG. In plants, 5mC occurs at CpG, CHG, and CHH sites! Luo developed DeepSignal-plant algorithms to address the multiple types of methylation in plants. Raw reads and a reference genome are needed as inputs. Tools used include Guppy and Tombo from ONT. DeepSignal-plant can detect methylation signals in all three sites. Results also indicated differences in methylation. They are refining the model and will continue using machine learning to improve the results. The output is a BED (bedmethyl) formatted file that contains methylation frequencies of called sites. Luo said that DeepSignal-plant “achieves high correlations with bisulfite sequencing in 5mC detection in CpG, CHG, CHH sites…” I would like to learn more about different sites and how the models are improved. I wonder if there are other methylation sites/contexts in bacterial genomes!
