Matrix factorization methods for integrative cancer genomics.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput sequencing generates multi-platform genomic data (e.g., DNA methylation, gene expression).
• Integrating these multidimensional genomic datasets presents unique opportunities and challenges for systems biology.
• Understanding coordination across regulatory levels is crucial for deciphering cellular systems.

Purpose of the Study:

• To present a matrix factorization framework for integrating multiple genomic datasets.
• To introduce a semi-supervised variant incorporating prior biological knowledge.
• To demonstrate the utility of this framework in uncovering complex biological patterns.

Main Methods:

• Matrix factorization framework to project diverse genomic data onto a common coordinate system.
• Identification of multidimensional modules where genetic variables are strongly correlated across samples.
• Application of a semi-supervised approach to enhance data integration with prior knowledge.

Main Results:

• The framework successfully integrates multi-platform genomic data, revealing coordinated regulatory patterns.
• Multidimensional modules highlight perturbed pathways and clinically distinct patient subgroups in cancer.
• Associations between distinct cellular activity layers are uncovered and their biological implications explained.

Conclusions:

• Matrix factorization provides a powerful approach for analyzing multidimensional genomic datasets.
• This method enhances the discovery of biological insights and patient stratification in complex diseases like cancer.
• Integrating diverse genomic data layers is essential for a comprehensive understanding of cellular systems.