A technical review of multi-omics data integration methods: from classical statistical to deep generative approaches

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• High-throughput sequencing generates large, complex multi-omics datasets.
• Multi-omics data integration is challenging due to high dimensionality, heterogeneity, and missing values.
• Computational methods are essential for analyzing multi-omics data to understand disease mechanisms.

Purpose of the Study:

• To comprehensively review state-of-the-art multi-omics integration methods.
• To focus on deep generative models, particularly variational autoencoders (VAEs).
• To outline future directions in precision medicine research through data integration.

Main Methods:

• Review of computational methods for multi-omics integration.
• Focus on deep generative models, including variational autoencoders (VAEs).
• Exploration of VAE loss functions, regularization techniques (adversarial training, disentanglement, contrastive learning), foundation models, and multimodal integration.

Main Results:

• Variational autoencoders (VAEs) are effective for data imputation, augmentation, and batch effect correction in multi-omics data.
• Advanced VAE techniques enhance the analysis of complex biological patterns.
• Foundation models and multimodal integration show promise for future research.

Conclusions:

• Deep generative models, especially VAEs, offer powerful solutions for multi-omics data integration challenges.
• These methods are critical for advancing precision medicine by improving disease mechanism understanding.
• Continued research in foundation models and multimodal integration will drive future discoveries.