MU-PseUDeep

MU-PseUDeep predicts pseudouridine (Ψ) sites in RNA sequences using deep learning to integrate sequence and predicted secondary-structure information for improved site identification.

Key Features:

• Deep learning architecture: Integrates raw RNA sequence data and predicted secondary structure as inputs to the model.
• Dual-input convolutional neural networks: Employs two sets of convolutional neural networks (CNNs) to capture sequence and structural contextual features.
• Comparative performance: Demonstrated improved pseudouridine site prediction accuracy relative to XG-PseU, PseUI, and iRNA-PseU across both balanced and imbalanced datasets.
• Transcriptome scanning: Applied to scan the human transcriptome to identify predicted Ψ sites.
• Functional enrichment findings: Genes with predicted Ψ sites were enriched for nucleotide and protein binding functions and neurodegeneration pathways.

Scientific Applications:

• Pseudouridine site prediction: Provides predictions of Ψ sites in RNA sequences for studies of RNA modification.
• RNA modification research: Supports investigation of pseudouridylation's roles in gene function and regulation.
• Transcriptome-wide functional analysis: Enables transcriptome-wide association of predicted Ψ sites with functional categories such as nucleotide/protein binding and neurodegeneration pathways.

Methodology:

Integrates raw RNA sequence and predicted secondary structure using a dual-input deep learning model composed of two convolutional neural network branches.