LPI-CSFFR

LPI-CSFFR predicts interactions between long non-coding RNAs (lncRNAs) and proteins using a convolutional neural network with serial feature fusion of sequences, secondary structures, and physicochemical properties to support discovery of molecular mechanisms and novel lncRNA–protein relationships.

Key Features:

• Novel Feature Fusion Method: Serially fuses features from sequences, secondary structures, and physicochemical properties to capture complex patterns in lncRNA–protein interactions.
• Deep Learning Architecture: Employs a convolutional neural network (CNN) with a feature reuse strategy to improve predictive performance.
• Benchmark Performance: Achieves 83.7% accuracy on RPI1460 and 98.1% accuracy on RPI1807 benchmark datasets.
• Cross-Species Generalization: Evaluated across five model organisms with Mus musculus reaching 99.5% prediction accuracy.
• Interaction Network Analysis: Constructs interaction networks to identify hotspot proteins involved in lncRNA–protein interactions.
• Prediction of Novel Interactions: Validated for predicting potential lncRNA–protein interactions in sample pairs with previously unknown interactions.

Scientific Applications:

• lncRNA functional analysis: Predicts lncRNA–protein interactions to aid studies of lncRNA roles in cellular processes.
• Molecular mechanism elucidation: Supports investigation of protein partners to clarify mechanisms underlying cellular functions and disease states.
• Cross-species comparative studies: Enables evaluation of lncRNA–protein interaction conservation and variation across multiple model organisms.
• Discovery of key proteins: Identifies hotspot proteins within interaction networks for targeted experimental follow-up.

Methodology:

Training of a CNN model using a serial feature fusion approach that integrates sequence, secondary structure, and physicochemical property features, incorporating a feature reuse strategy and constructing interaction networks for hotspot identification.