Statistical integration of two omics datasets using GO2PLS.

Area of Science:

• Multi-omics data integration
• Systems biology
• Bioinformatics

Background:

• Biological systems are studied using multiple omics datasets, necessitating data integration for comprehensive understanding.
• Existing methods like Partial Least Squares (PLS) and Orthogonal PLS (O2PLS) have limitations in handling data heterogeneity and identifying sparse feature subsets.

Purpose of the Study:

• To develop a novel method, Group Sparse O2PLS (GO2PLS), for integrating two omics datasets.
• To enhance feature selection by incorporating biological group structures for improved interpretability.

Main Methods:

• Extended O2PLS to incorporate group structures among variables, enabling group selection in the joint subspace.
• Developed Group Sparse O2PLS (GO2PLS) to achieve sparsity and improve feature selection performance.

Main Results:

• Simulation studies demonstrated that GO2PLS improves accuracy in joint score and loading estimation and enhances feature selection.
• Application to TwinsUK and CVON-DOSIS datasets identified biologically relevant features related to the immune system and heart muscle disease, respectively.

Conclusions:

• GO2PLS effectively integrates multi-omics data by leveraging external group information.
• The method provides a sparse subset of features, enhancing interpretability and understanding of biological systems.