A consensus orthogonal partial least squares discriminant analysis (OPLS-DA) strategy for multiblock Omics data...

Area of Science:

• Systems Biology
• Bioinformatics
• Analytical Chemistry

Background:

• Omics approaches offer broad biological system monitoring but no single technique captures full biochemical content.
• Complex biological data require fusion of information from multiple sources for comprehensive analysis.
• High-dimensional omics data present challenges in knowledge extraction.

Purpose of the Study:

• To develop a generic methodology for fusing omics data from multiple sources.
• To provide an efficient tool for handling high-dimensional and noisy omics datasets.
• To assess the potential of the proposed fusion method through real-world case studies.

Main Methods:

• A novel methodology combining multiple kernel learning and Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA).
• Application of the fusion method to diverse omics datasets, including metabolomics, transcriptomics, and proteomics.
• Validation using three case studies: plant metabolomics, grape variety classification, and cancer cell line data analysis.

Main Results:

• Demonstrated successful fusion of mass spectrometry-based metabolomic data from different ionization modes.
• Successfully classified wine grape varieties using 2D heteronuclear magnetic resonance spectroscopy data.
• Effectively combined heterogeneous systems biology data, including publicly available NCI-60 cancer cell line omics data.

Conclusions:

• The proposed method is a relevant and widely applicable tool for efficient fusion of multiple omics data.
• Integrating omics data from different sources provides a more complete view of biological systems.
• The method effectively handles the inherent characteristics of multiple omics data, such as noisy collinear variables.