Prof. Liu Zhendong from the School of Computer and Information Engineering, in collaboration with Prof. Xiao Chuanle’s team at the State Key Laboratory of Ophthalmology, Sun Yat-sen University, has published a paper titled Accurate cross-species 5mC detection for Oxford Nanopore sequencing in plants with DeepPlant in the prestigious journal Nature Communications (a Chinese Academy of Sciences (CAS) Top journal, with an Impact Factor of 14.7 in 2024). This marks the first time our institute has achieved such a milestone in publishing in a top-tier Nature journal.

The paper introduces DeepPlant, a deep learning model that integrates bidirectional long short-term memory networks (Bi-LSTM) and Transformer architectures. This model significantly improves the detection accuracy of CHH sites, constructs a training and testing dataset covering diverse 9-mer motifs, and effectively addresses the scarcity of CHH methylation-positive training samples.

In evaluations across nine plant species, DeepPlant achieved a correlation coefficient of 0.705–0.838 between predicted genome-wide methylation frequencies at CHH sites and bisulfite sequencing data, representing a 23.4%–117.6% improvement over the existing tool Dorado.

The model also demonstrates exceptional single-molecule detection accuracy and F1 scores, providing a powerful, broadly applicable analytical tool for plant epigenetic research.