SDGCCA: Supervised Deep Generalized Canonical Correlation Analysis for Multi-Omics Integration

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics
• Systems Biology

Background:

• Integrating multi-omics data is crucial for understanding complex biological mechanisms and phenotypes.
• Existing canonical correlation analysis (CCA) methods have limitations in modeling nonlinear, multi-view correlations and performing phenotype classification.
• A need exists for advanced methods that can handle complex cross-data correlations across multiple biological data types.

Purpose of the Study:

• To introduce a novel multi-omics integration method, supervised deep generalized canonical correlation analysis (SDGCCA).
• To model nonlinear correlation structures between multiple omics data modalities for enhanced phenotype classification.
• To identify multi-omics biomarkers associated with specific phenotypes.

Main Methods:

• Developed SDGCCA, a nonlinear, multi-view CCA projection method.
• SDGCCA models complex/nonlinear cross-data correlations between multiple ($$2) omics modalities.
• The method performs phenotype classification and feature ranking.

Main Results:

• SDGCCA outperformed existing CCA-based and supervised methods in predicting Alzheimer's disease (AD) and discriminating early- and late-stage cancers.
• Demonstrated SDGCCA's capability for feature selection to identify important multi-omics biomarkers.
• Applied to AD data, SDGCCA identified known AD-associated gene clusters within multi-omics data.

Conclusions:

• SDGCCA is an effective nonlinear multi-view method for multi-omics integration, phenotype classification, and biomarker discovery.
• The method advances the field by addressing limitations of previous CCA-based approaches, particularly in handling multiple data views and nonlinearities.
• SDGCCA shows significant potential for applications in precision medicine and understanding complex diseases.