MODAPro: Explainable Heterogeneous Networks with Variational Graph Autoencoder for Mining Disease-Specific Functional...

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Multiomics data integration faces challenges due to heterogeneity, sparsity, and interpretability gaps.
• Existing analytical methods struggle to capture complex, nonlinear molecular relationships across omics layers.

Purpose of the Study:

• To introduce MODAPro, a deep learning framework for effective multiomics data integration.
• To address limitations in current approaches for disease mechanism investigation.

Main Methods:

• MODAPro synergistically integrates variational graph autoencoders (VAE) with graph convolutional networks (GCN).
• The framework employs a biologically informed deep learning architecture.

Main Results:

• MODAPro outperforms existing methods in identifying disease-associated biomarkers and functionally coherent modules.
• It reveals latent biomolecular information missed by conventional techniques.
• The framework captures intricate across-omic interactions, enhancing functional annotation and providing systems-level insights.

Conclusions:

• MODAPro offers a robust approach for multiomics integration, advancing systems biology and translational medicine.
• Its adaptability supports precision medicine by uncovering actionable disease signatures and regulatory networks.
• The framework facilitates discovery even with sparse or incomplete single-omics data.