SMODA: Interpretable Multimodal Omics Integration for Disease Classification and Subtype Discovery via Heterogeneous...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Disease heterogeneity complicates precision medicine development.
• Existing multimodal data integration methods struggle with omics noise, data imbalance, and interpretability.

Purpose of the Study:

• To introduce SMODA (Semi-Supervised Multimodal Omics Data Analysis), a flexible framework for integrating multimodal omics data.
• To address limitations of current methods by reducing cross-modal heterogeneity and improving interpretability.

Main Methods:

• SMODA combines heterogeneous transfer learning and semisupervised modeling.
• It learns shared latent representations across different data modalities.
• The framework was systematically benchmarked against existing multiomics integration methods.

Main Results:

• SMODA demonstrated superior performance in disease classification and subtype identification compared to existing methods.
• Application to esophageal cancer data confirmed improved classification and identified a novel, clinically relevant subtype.
• The newly identified subtype exhibits distinct metabolic, inflammatory, and exposure features associated with poor prognosis.

Conclusions:

• SMODA offers a reliable and interpretable framework for multimodal omics data integration.
• The method supports clinically relevant disease stratification and advances precision medicine.
• SMODA facilitates the discovery of novel disease subtypes and their underlying biological mechanisms.