MVDA

MVDA integrates multi-view genomic and clinical data (mRNA expression, miRNA expression, DNA methylation, and clinical data) to improve patient subtype classification in multi-omics studies.

Key Features:

• Multi-View Integration: Employs a late integration approach that factorizes membership matrices to combine results from individual view clustering iterations.
• Dimension Reduction and Variable Selection: Incorporates dimension reduction and variable selection techniques to retain the most relevant features for analysis.
• Per-View Clustering and Integration: Performs clustering separately for each data type (view) before integrating those clustering results.
• Quantification of View Contributions: Quantifies the contribution of individual views to the final integrated subtype assignments.
• Statistical Validation and Novel Subgroup Identification: Has been evaluated on six multi-view cancer datasets, identifying statistically significant patient subclasses and novel sub-groups relative to other multi-view clustering methods.
• Integration of Prior Information: Supports incorporation of prior information alongside genomic features in subtyping analyses.

Scientific Applications:

• Patient subclassification in oncology: Enhances identification of clinically relevant patient subtypes by integrating mRNA, miRNA, DNA methylation, and clinical data.
• Discovery of novel disease subgroups: Enables detection of previously underrepresented or novel cancer sub-groups with statistical support.
• Assessment of omics contributions: Provides quantitative assessment of how different molecular profiles influence subtype definitions.
• Insights into mechanisms and targets: Supports generation of hypotheses about disease mechanisms and potential therapeutic targets through integrated multi-omics analysis.

Methodology:

Uses a late integration strategy that factorizes membership matrices derived from per-view clustering iterations, combined with dimension reduction, variable selection, quantification of view contributions, and optional integration of prior information.