A robust deep learning framework for RNA 5-methyluridine modification prediction using integrated features.

Area of Science:

• Computational Biology
• Molecular Biology
• Bioinformatics

Background:

• RNA 5-methyluridine (m5U) modifications are crucial for various biological processes.
• Accurate prediction of m5U sites is vital for understanding gene regulation and function.

Purpose of the Study:

• To develop a highly accurate computational tool, 5-meth-Uri, for predicting RNA 5-methyluridine (m5U) modifications.
• To improve upon existing methods for m5U site prediction.

Main Methods:

• Feature extraction using dinucleotide and trinucleotide auto-cross covariance with physicochemical parameters.
• Dimensionality reduction via Principal Component Analysis (PCA).
• Classification using a Deep Neural Network (DNN).

Main Results:

• The 5-meth-Uri model achieved high prediction accuracy, reaching up to 97.36% on training data and 96.51% on independent test sets.
• Demonstrated significant accuracy improvements (up to 7.15%) over existing models on benchmark datasets.
• Validated performance using tenfold cross-validation and independent testing on mature mRNA and full transcript datasets.

Conclusions:

• 5-meth-Uri is a highly accurate and reliable tool for predicting RNA m5U modifications.
• The tool has potential applications in biomedical research and pharmaceutical development.
• The developed method advances computational approaches for analyzing RNA modifications.