DSGCNLDA: A Multi-view Learning Model with DualScope Attention for lncRNA-Disease Association Prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Long non-coding RNAs (lncRNAs) are crucial regulators of biological processes.
• Accurate prediction of lncRNA-disease associations aids in understanding disease mechanisms and developing treatments.
• Current computational methods struggle with data sparsity, incompleteness, and insufficient node representation.

Purpose of the Study:

• To propose a novel computational model, DSGCNLDA, for enhanced prediction of lncRNA-disease associations.
• To address limitations of existing methods by improving data representation and capturing complex network topology.

Main Methods:

• Multi-view fusion learning to integrate diverse biological similarity features into a comprehensive similarity matrix.
• Construction of a heterogeneous network of lncRNA-disease associations using similarity and adjacency matrices.
• Feature extraction via a graph convolutional network (GCN) encoder with a novel DualScope attention mechanism for node representation.
• Association prediction using a multi-layer perceptron (MLP).

Main Results:

• DSGCNLDA demonstrated strong performance in predicting lncRNA-disease associations across multiple public datasets.
• Ablation studies validated the novelty and contribution of the proposed components.
• Case studies and generalization evaluations confirmed the model's effectiveness in biomedical prediction.

Conclusions:

• The DSGCNLDA model effectively enhances lncRNA-disease association prediction by leveraging multi-view learning and an advanced graph convolutional network with a DualScope attention mechanism.
• The proposed approach overcomes data limitations and improves the representation of complex biological networks.
• DSGCNLDA shows significant potential for applications in disease pathophysiology research and therapeutic strategy development.