Heterogeneous dual-channel and interpretable graph representation learning with global virtual nodes for...

Area of Science:

• Computational biology
• Pharmacogenomics
• Bioinformatics

Background:

• MicroRNAs (miRNAs) regulate drug response, impacting therapeutic outcomes.
• Existing computational methods struggle with data sparsity and heterogeneous networks for miRNA-drug sensitivity prediction.

Purpose of the Study:

• To develop an advanced computational framework, HDIGRL, for accurate miRNA-mediated drug sensitivity prediction.
• To address limitations of existing methods in handling complex network structures and data sparsity.

Main Methods:

• Proposed HDIGRL, a channel-aware heterogeneous graph representation learning framework.
• Utilized dual-channel feature extraction and channel-gated global heterogeneous propagation.
• Incorporated an imbalance-aware focal loss for improved robustness.

Main Results:

• HDIGRL demonstrated superior performance over existing methods on public datasets.
• The framework effectively propagated features across heterogeneous network structures.
• Identified potential miRNA-mediated drug-sensitivity pathways.

Conclusions:

• HDIGRL offers a robust and effective approach for predicting miRNA-mediated drug sensitivity.
• The framework has potential for both predictive modeling and biological interpretation in drug discovery.