# PlantDeepMeth: A Deep Learning Model for Predicting DNA Methylation States in Plants

## TL;DR

PlantDeepMeth is a deep learning model that predicts DNA methylation states in plants, showing good performance and generalizability across species.

## Contribution

A novel deep learning model for predicting plant DNA methylation states with cross-species generalizability.

## Key findings

- PlantDeepMeth performs well in predicting methylation states in Brassica rapa and Arabidopsis thaliana.
- Motif analysis reveals specific patterns linked to hypo- or hyper-methylation in the studied plant species.
- The model maintains high performance across different plant species, showing generalizability.

## Abstract

Cytosine DNA methylation (5mCs) is an important epigenetic modification in genomic research. However, the methylation states of some cytosine sites are not available due to the limitations of different studies, and there are few tools developed to deal with this problem, especially in plants, which have more methylation types than animals. Here, we report PlantDeepMeth, a novel deep learning model that utilizes deep learning to predict DNA methylation states in plants. The evaluation of PlantDeepMeth on known cytosine sites in both the Brassica rapa and Arabidopsis thaliana genomes shows good performance in predicting methylation states, indicating that the tool is good at learning patterns for methylation imputation. Motif analysis of the model’s predictions identified specific motifs associated with hypo- or hyper-methylation states in B. rapa and A. thaliana, further revealing key regulatory patterns captured by the model. Moreover, cross-species validation between B. rapa and A. thaliana demonstrated the generalizability of PlantDeepMeth, with the model maintaining high performance across different plant species. These results highlight the effectiveness of PlantDeepMeth and demonstrate the potential of deep learning to advance plant genomics research.