A unified framework for correcting batch effects and integrating multi-omics data.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Multi-omics studies integrate diverse molecular data (gene expression, DNA methylation, chromatin accessibility) for comprehensive biological insights.
• Integrating heterogeneous public datasets introduces significant batch effects and technical variability, complicating analysis.
• Existing batch correction methods are often limited to single-omics data, failing to address multi-omics integration challenges.

Purpose of the Study:

• To develop a unified framework, MoDAmix, for effective multi-omics batch effect correction and integration.
• To harmonize heterogeneous datasets while preserving shared biological structure across different omics layers.
• To enable reliable cross-cohort analysis for systems biology and precision medicine.

Main Methods:

• MoDAmix employs domain adaptation and adversarial learning to align feature distributions across batches and modalities.
• The framework involves pre-training, within-omics adversarial adaptation, multi-omics adversarial alignment, and semi-supervised class alignment.
• It enforces consistency within and between omics types to achieve coherent cross-omics integration in a shared latent space.

Main Results:

• MoDAmix effectively mitigates batch effects in both single-cell and bulk multi-omics datasets.
• The framework demonstrated improved clustering and classification performance across different biological domains.
• MoDAmix successfully preserved essential subtype structures, proving its robustness in harmonizing heterogeneous data.

Conclusions:

• MoDAmix provides a robust solution for multi-omics batch effect correction and integration.
• The framework facilitates reliable cross-cohort analysis, advancing systems biology and precision medicine.
• MoDAmix is publicly available, promoting its adoption in biological research.