A multivariate approach to the integration of multi-omics datasets

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics
• Proteomics

Background:

• Multi-omics studies require advanced exploratory data analysis for systematic integration and comparison of multiple omics layers.
• Multiple co-inertia analysis (MCIA) is introduced as a method to identify co-relationships between high-dimensional datasets.
• MCIA projects multiple datasets into the same dimensional space for enhanced biological interpretation and pathway analysis.

Purpose of the Study:

• To describe and demonstrate the utility of Multiple Co-Inertia Analysis (MCIA) for multi-omics data integration.
• To showcase MCIA's ability to identify co-relationships and extract meaningful biological insights from diverse omics datasets.
• To apply MCIA in real-world scenarios, including cancer cell line analysis and tumor molecular subtyping.

Main Methods:

• Multiple Co-Inertia Analysis (MCIA) based on a covariance optimization criterion.
• Simultaneous projection of multiple datasets into a common dimensional space.
• Application of MCIA to integrate transcriptome and proteome profiles (NCI-60 cancer cell lines) and compare transcriptome data from different platforms (ovarian tumors).

Main Results:

• Integration of transcriptome and proteome data revealed complementary features, enhancing pathway analysis and highlighting the leukemia extravasation signaling pathway.
• Comparison of transcriptome profiles from different platforms identified RNA-sequencing data (RPKM) as having greater variance.
• Novel biomarkers associated with tumor molecular subtypes were identified by combining data from four platforms.

Conclusions:

• MCIA is an effective method for data integration and visualization of multi-omics datasets from the same individuals.
• The method's independence from feature annotation allows for the extraction of important features not present across all datasets.
• MCIA offers simple graphical representations for identifying relationships within large, complex datasets.