MU-PseUDeep: A deep learning method for prediction of pseudouridine sites

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Pseudouridine (Ψ) is a modified nucleotide crucial for RNA function.
• Pseudouridine synthase catalyzes the conversion of uridine to pseudouridine.
• Existing machine-learning methods often rely on nucleotide frequency, potentially missing conformational context.

Purpose of the Study:

• To develop a more accurate computational tool for predicting pseudouridine sites.
• To leverage deep learning for capturing both sequence and secondary structure information.
• To investigate the functional implications of predicted pseudouridine sites in the human transcriptome.

Main Methods:

• Developed MU-PseUDeep, a deep learning tool utilizing raw RNA sequence and predicted secondary structure.
• Employed two sets of convolutional neural networks to process sequence and structure features.
• Applied MU-PseUDeep to scan the human transcriptome for pseudouridine sites.

Main Results:

• MU-PseUDeep demonstrated significant improvements in pseudouridine site prediction accuracy compared to existing tools (XG-PseU, PseUI, iRNA-PseU).
• The tool performed well on both balanced and imbalanced datasets.
• Analysis of the human transcriptome revealed enrichment of predicted pseudouridine sites in genes involved in nucleotide/protein binding and neurodegeneration pathways.

Conclusions:

• MU-PseUDeep represents a state-of-the-art tool for pseudouridine site prediction.
• The findings highlight the importance of sequence and secondary structure context in pseudouridine modification.
• Predicted pseudouridine sites are associated with key biological processes and pathways, including neurodegeneration.