JLCRB

JLCRB predicts RNA-binding protein (RBP) binding sites on circular RNAs (circRNAs) using multi-view joint representation learning to integrate and calibrate diverse feature encodings.

Key Features:

• Multi-View Feature Encoding: Utilizes diverse encoding methods to extract comprehensive RNA-related features from multiple perspectives.
• Intrinsic Connection Construction: Constructs a global joint representation that captures intrinsic relationships between different view features rather than simple concatenation.
• Feature Calibration: Calibrates each view's features based on the global joint representation to emphasize critical features and suppress less relevant ones.
• Depth Feature Fusion: Integrates depth features derived from multiple views and fuses them for accurate detection of RBP binding sites on circRNAs.

Scientific Applications:

• CircRNA–RBP interaction mapping: Enables identification of RBP binding sites on circRNAs to support studies of RNA–protein interactions.
• Gene regulation and disease research: Facilitates investigation of circRNA roles in gene regulation and disease mechanisms by providing binding-site information.
• Performance benchmark: Demonstrated an average Area Under the Curve (AUC) of 93.68% across 37 CircRNA-RBP datasets.

Methodology:

Feature extraction via multiple encoding methods; construction of a global joint representation by integrating multi-view features; calibration of view-specific features using the global joint representation; fusion of depth features for binding site detection.