Multi-Head Hypergraph Convolution With Feature Enhancement and Latent Representation Learning for miRNA-Disease...

Area of Science:

• Bioinformatics
• Computational Biology
• Network Medicine

Background:

• MicroRNA (miRNA)-disease associations are vital for understanding disease mechanisms.
• Existing computational methods struggle with sparse data and capturing deep topological structures.

Purpose of the Study:

• To develop a novel framework, FKAMHV, for robust miRNA-disease association prediction.
• To enhance the extraction of deep topological features and uncover latent associations, especially under sparse conditions.

Main Methods:

• Constructed heterogeneous networks and generated miRNA/disease-specific hypergraphs.
• Integrated Fast Kolmogorov-Arnold Networks (FastKAN) for nonlinear feature modeling and Multi-Head Hypergraph Convolutional Networks (Multi-Head HGCN) for joint representation.
• Employed $\beta$-Variational Autoencoder ($\beta$-VAE) for latent association modeling and introduced attention mechanisms and Jumping Knowledge strategy within HGCN.

Main Results:

• FKAMHV demonstrated superior performance over existing methods in terms of Area Under the Curve (AUC) and Area Under the Precision-Recall Curve (AUPR).
• The framework achieved strong predictive performance even with sparse association data.

Conclusions:

• FKAMHV effectively captures complex topological structures and latent associations for miRNA-disease prediction.
• The proposed method offers improved generalization and robustness, particularly in sparse data settings, advancing the field of computational disease association studies.