Using a hybrid neural network architecture for DNA sequence representation: A study on N⁴-methylcytosine sites

Area of Science:

• Genomics and Epigenetics
• Bioinformatics and Computational Biology
• Plant Science

Background:

• N4-methylcytosine (4mC) is a crucial DNA modification involved in epigenetic regulation across various plant genomes.
• The Rosaceae family, including important fruit crops, exhibits 4mC modifications with implications for gene expression, adaptation, and evolution.
• Accurate prediction of 4mC sites is vital for understanding its functional roles in plant development and stress responses.

Purpose of the Study:

• To develop a highly accurate computational model for predicting 4mC sites in the Fragaria vesca genome.
• To improve upon existing prediction methods by integrating advanced feature encoding and deep learning architectures.
• To provide a robust tool for epigenetic research in Rosaceae species.

Main Methods:

• Utilized deep learning, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long Short-Term Memory (LSTM) networks.
• Incorporated advanced feature encoding techniques and pre-trained natural language processing (NLP) models, such as fastText.
• Evaluated model performance using sensitivity, specificity, and accuracy on an independent dataset.

Main Results:

• The best performing model was a CNN architecture employing fastText encoding.
• This model achieved high predictive performance with a sensitivity of 0.909, specificity of 0.77, and accuracy of 0.879.
• The developed model demonstrated superior predictive capabilities compared to previously published methods on the same dataset.

Conclusions:

• The enhanced deep learning model significantly improves the accuracy of N4-methylcytosine (4mC) site prediction in Fragaria vesca.
• This advancement offers a valuable tool for epigenetic studies in Rosaceae plants, aiding research into gene regulation and adaptation.
• The findings highlight the potential of integrating NLP techniques with deep learning for precise epigenetic marker identification in plant genomes.