PseUdeep

PseUdeep predicts pseudouridine (Ψ) modification sites in RNA sequences from Homo sapiens, Saccharomyces cerevisiae, and Mus musculus to support investigation of RNA modification roles in biological processes.

Key Features:

• Deep learning framework: Employs an advanced deep learning framework to predict pseudouridine (Ψ) sites.
• Feature extraction methods: Extracts sequence features using one-hot encoding, K-tuple nucleotide frequency pattern, and position-specific nucleotide composition.
• Neural network architecture: Applies two convolutional operations on extracted feature matrices and processes them with a capsule neural network and a bidirectional gated recurrent unit (GRU) network enhanced by a self-attention mechanism for classification.
• Dimensionality reduction: Reduces feature dimensionality to 109,109 for H. sapiens, 119 for M. musculus, and 119 for S. cerevisiae.

Scientific Applications:

• Disease mechanism elucidation: Facilitates study of molecular mechanisms underlying diseases by locating Ψ sites in RNA.
• RNA modification research: Enables investigation of biological processes influenced by pseudouridine modifications.
• Experimental alternative: Provides a computational approach to reduce reliance on laboratory techniques for Ψ site identification.

Methodology:

Sequence features are encoded by one-hot, K-tuple nucleotide frequency, and position-specific composition; feature matrices undergo two convolutional operations and are processed by a capsule neural network and a bidirectional GRU with self-attention; dimensionality is reduced to species-specific sizes and performance was evaluated by tenfold cross-validation and independent testing on S-200 and H-200 datasets.