circRB

circRB identifies sequence specificities of circular RNA (circRNA)-binding proteins using a capsule network-based model to characterize circRNA–RBP interactions.

Key Features:

• Capsule Network Architecture: Employs a variant of the capsule network architecture to capture sequence features of circRNAs.
• Convolutional Feature Extraction: Utilizes convolution operations to extract detailed sequence features from input sequences.
• Two Dynamic Routing Algorithms: Processes convolution-derived features through two dynamic routing algorithms within the capsule network to discriminate binding sites.
• Classification Framework: Integrates convolution operations and dynamic routing within a classification framework to assign sequence specificities.
• Performance Superiority: Demonstrates higher prediction accuracy and outperforms existing computational methods in identifying sequence specificities of circRNA–RBP interactions.

Scientific Applications:

• Detection of Sequence Motifs: Detects sequence motifs across seven circRNA–RBP bound sequence datasets and matches them with known human RNA motifs.
• Overlap Analysis: Identifies overlaps between motifs on circular RNAs and those on linear RNAs for comparative analysis.
• Prediction of Binding Sites: Predicts binding sites on reported full-length sequences of circRNAs interacting with RBPs to provide candidates for experimental validation.

Methodology:

Uses a classification framework that combines convolution operations and a capsule network with two dynamic routing algorithms to analyze sequence features.