Capsule-LPI

Capsule-LPI predicts long noncoding RNA-protein interactions (LPIs) by integrating multimodal sequence, motif, physicochemical, and secondary-structure features into a multichannel capsule network to support inference of lncRNA functional roles.

Key Features:

• Multichannel capsule network framework: Uses a multichannel capsule network architecture to model complex relationships in LPI prediction.
• Multimodal feature integration: Integrates four distinct feature groups—sequence features, motif information, physicochemical properties, and secondary structure features—for prediction.
• Dedicated feature-learning subnetworks: Processes each feature group through separate feature-learning subnetworks prior to integration.
• Capsule subnetwork integration: Combines outputs of feature-learning subnetworks into a cohesive capsule subnetwork for final prediction.
• Deep learning approach: Employs deep learning techniques to enhance representation learning for LPIs.
• Improved predictive performance: Demonstrated increased metrics with precision of 87.3% (1.7% improvement) and F-value of 92.2% (1.4% improvement) compared to prior methods.

Scientific Applications:

• lncRNA functional inference: Supports elucidation of lncRNA functional roles by predicting their protein interaction partners.
• Gene regulation studies: Facilitates investigation of lncRNA-mediated regulatory mechanisms through predicted LPIs.
• Biological process characterization: Aids study of biological processes involving lncRNAs by providing candidate lncRNA-protein interactions.

Methodology:

Processes four feature groups (sequence, motif, physicochemical, secondary structure) via dedicated feature-learning subnetworks and integrates them using a multichannel capsule network and a capsule subnetwork within a deep learning framework.