SPARSE INTEGRATIVE CLUSTERING OF MULTIPLE OMICS DATA SETS.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-resolution genomic profiling (microarrays, sequencing) reveals disease-associated DNA copy number, methylation, and gene expression alterations.
• Integrated genomic profiling of multiple omics data types in the same samples offers superior resolution compared to single data types.

Purpose of the Study:

• To develop and validate a penalized latent variable regression method for joint modeling of multiple omics data types.
• To identify common latent variables for clustering patient samples into biologically and clinically relevant disease subtypes.
• To apply the integrated approach for subtype analysis in breast and lung cancer datasets.

Main Methods:

• Utilized penalized latent variable regression, including lasso, elastic net, and fused lasso, to induce sparsity and identify key genomic features.
• Employed iterative ridge regression for computing sparse coefficient vectors.
• Incorporated uniform design for efficient tuning parameter selection and model comparison against sparse Singular Value Decomposition (SVD) and penalized Gaussian Mixture Model (GMM).

Main Results:

• The proposed method successfully integrates genomic, epigenomic, and transcriptomic data.
• Demonstrated the ability to cluster patient samples into distinct disease subtypes.
• Outperformed sparse SVD and penalized GMM in analyses of both simulated and real-world cancer datasets.

Conclusions:

• Penalized latent variable regression provides a powerful framework for integrated multi-omics data analysis.
• This approach facilitates the discovery of novel disease subtypes with potential clinical and biological significance.
• The method shows promise for advancing precision medicine through comprehensive genomic profiling.