DeepR2OM: Accurate Recognition for RNA 2'-O-Methylation Sites in Human Genome Using Deep Learning.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• 2'-O-methylation (2OM) is a key RNA modification affecting RNA stability, structure, and function.
• Predicting 2OM sites is vital for understanding RNA biology and associated diseases.
• Existing detection methods are often resource-intensive, costly, and can damage RNA samples.

Purpose of the Study:

• To develop a novel, accurate, and efficient computational method for predicting RNA 2OM sites.
• To leverage machine learning, specifically deep learning, for rapid and cost-effective 2OM site prediction.
• To provide an accessible web server tool for RNA 2OM site prediction.

Main Methods:

• Developed DeepR2OM, integrating RNA sequence encoding with feature selection and deep learning.
• Utilized eight RNA descriptors and feature selection algorithms for dimensionality reduction.
• Employed Convolutional Neural Network (CNN), Multi-Head Self-Attention, and Deep Neural Network (DNN) models.

Main Results:

• DeepR2OM achieved high performance on an independent test set: 87.1% accuracy, 85.5% recall, 87.9% precision, and 75.7% MCC.
• The selected deep learning models (CNN, Multi-Head Self-Attention, DNN) demonstrated effectiveness in 2OM site prediction.
• Experimental validation confirmed the method's predictive power.

Conclusions:

• DeepR2OM offers a robust and efficient computational approach for predicting RNA 2OM sites.
• The developed tool aids in exploring the functional and bioinformatic significance of RNA methylation.
• An accessible web server is available for broader research community use.