Transformer‐based DNA methylation detection on ionic signals from Oxford Nanopore sequencing data

Abstract: DNA methylation is a key epigenetic modification with important implications for gene regulation and disease. Oxford Nanopore sequencing enables direct detection of methylation from raw ionic current signals, offering advantages over traditional bisulfite sequencing and array-based methods. In this study, we present a novel application of Transformer architecture—a deep learning model based on self-attention mechanisms—for the detection of DNA methylation from Nanopore signal data. Leveraging its ability to capture long-range dependencies and focus on contextually important bases, the Transformer model was trained and evaluated on real datasets, including Escherichia coli and the human genome sample NA12878. Our results demonstrate that the Transformer outperforms traditional deep learning models such as convolutional and recurrent neural networks in methylation detection accuracy and sequence context interpretation. This work highlights the potential of attention-based models for advancing epigenomic analysis from long-read sequencing technologies.

Xiuquan Wang, Mian Umair Ahsan, Yunyun Zhou, Kai Wang. Transformer-based DNA methylation detection on ionic signals from Oxford Nanopore sequencing data. Quant. Biol., 2023, 11(3): 287‒296.