NPI-RGCNAE

NPI-RGCNAE predicts interactions between non-coding RNAs (ncRNAs) and RNA-binding proteins to identify ncRNA–protein interaction networks.

Key Features:

• Relational Graph Convolutional Network (R-GCN) Encoder: Employs an R-GCN encoder to model complex relationships between ncRNAs and proteins and capture structural patterns in the interaction graph.
• DistMult Decoder: Uses a DistMult decoder to reconstruct the interaction matrix from encoded latent representations for interaction prediction.
• Efficient Negative Sample Selection Strategy: Implements a strategy for selecting negative samples to improve robustness during model training and validation.
• Performance and Efficiency: Demonstrated comparable performance to state-of-the-art methods via 5-fold cross-validation while requiring <10% of the training time of other leading approaches.

Scientific Applications:

• Regulatory network analysis: Predicts ncRNA–protein interactions to support mapping of regulatory roles of ncRNAs in gene expression.
• Disease mechanism investigation: Supports studies linking ncRNA–protein interactions to disease mechanisms.
• Cellular process characterization: Facilitates identification of ncRNA partners involved in cellular processes.
• High-throughput screening: Suitable for large-scale prediction of ncRNA–protein interactions in high-throughput datasets.
• Integrative systems biology analyses: Enables incorporation of predicted interactions into systems-level analyses.

Methodology:

Constructs a relational graph of known ncRNA–protein interactions, encodes graph structure with an R-GCN encoder, reconstructs potential interaction pairs with a DistMult decoder, and uses an efficient negative sample selection strategy and 5-fold cross-validation for evaluation.