iCircRBP-DHN

iCircRBP-DHN identifies RNA-binding protein (RBP) binding sites on circular RNAs (circRNAs) using a deep hierarchical network and novel encoding schemes to enable genome-wide characterization of circRNA–RBP interactions.

Key Features:

• Deep Hierarchical Network Architecture: A deep multi-scale residual network combined with bidirectional gated recurrent units (BiGRUs) and a self-attention mechanism extracts local and global contextual information from RNA sequences.
• Novel Encoding Schemes: Integrates CircRNA2Vec and k-tuple nucleotide frequency patterns to capture various degrees of nucleotide dependencies for discriminative feature representation.
• Validation and Performance: Validated on 37 circRNA datasets and 31 linear RNA datasets, demonstrating consistent performance improvements over state-of-the-art algorithms.
• Motif Analysis: Performs motif analysis on circRNAs bound by different RBPs to reveal sequence motifs associated with binding.
• Numerical Stability and Scalability: Architecture and encoding choices address numerical instability and scalability issues reported in prior methods.

Scientific Applications:

• CircRNA–RBP Binding Site Identification: Accurately identifies RBP binding sites on circRNAs for genome-wide interaction mapping.
• Regulatory Mechanism Exploration: Facilitates study of circRNA involvement in gene expression regulation by mapping RBP interactions.
• Disease and Cellular Process Investigation: Supports investigation of cellular processes and disease pathogenesis involving circRNAs through interaction and motif analyses.

Methodology:

Processes cross-linking immunoprecipitation with sequencing data, encodes sequences using CircRNA2Vec and k-tuple nucleotide frequencies, applies a deep multi-scale residual network with BiGRUs and self-attention, and conducts motif analysis.