CRBPDL

CRBPDL predicts binding sites between circular RNAs (circRNAs) and RNA-binding proteins (RBPs) to characterize posttranscriptional regulatory interactions.

Key Features:

• Adaboost-integrated deep hierarchical network: Combines an Adaboost algorithm with a deep hierarchical network architecture to improve robustness and reliability of predictions.
• Five feature encoding schemes: Transforms original RNA sequences using five distinct encoding schemes to capture diverse biological and structural information.
• Deep multiscale residual networks (MSRN): Uses MSRN to extract multi-resolution local features from encoded sequence representations.
• Bidirectional gated recurrent units (BiGRUs): Employs BiGRUs to model global contextual and sequential dependencies in RNA sequences.
• Self-attention mechanism: Applies self-attention to emphasize relevant features within sequence representations for improved prediction accuracy.

Scientific Applications:

• Identification of circRNA-RBP binding sites: Predicts interaction sites to support studies of circRNA-mediated posttranscriptional regulation.
• Functional and disease-related studies: Provides candidate interaction sites to aid elucidation of circRNA roles in biological processes and diseases.
• Cross-dataset validation: Demonstrated performance across 37 circular RNA datasets and 31 linear RNA datasets for broad applicability.

Methodology:

Integrates an Adaboost algorithm with a deep hierarchical network, employs five feature encoding schemes on original RNA sequences, uses deep multiscale residual networks (MSRN) and BiGRUs to learn representations, and incorporates a self-attention mechanism.