mixedCCA

mixedCCA performs sparse canonical correlation analysis (CCA) using latent Gaussian copula models to estimate dependencies between continuous, binary, and zero-inflated variables for multi-view data integration and is implemented as an R package.

Key Features:

• Integration of mixed variable types: Uses latent Gaussian copula models to represent dependencies among continuous, binary, and zero-inflated variables.
• Efficient computational approach: Estimates latent correlations with a hybrid multilinear interpolation and optimization scheme to accelerate computation.
• Scalability and performance: Applies to high-dimensional settings where the number of variables p is large, reports speed-ups of several orders of magnitude relative to existing methods, and provides theoretical guarantees on approximation error.
• Demonstrated applications: Validated on simulated data and applied to high-dimensional sparse quantitative and relative abundance microbiome data and multi-view datasets from The Cancer Genome Atlas Project.

Scientific Applications:

• Microbiome research: Analysis and integration of high-dimensional, sparse quantitative and relative abundance microbiome data comprising mixed variable types.
• Cancer genomics: Integration and exploration of multi-view genomic datasets from The Cancer Genome Atlas Project to study relationships across data types.

Methodology:

Estimation of latent Gaussian correlations via a hybrid multilinear interpolation and optimization approach to perform sparse CCA, with theoretical approximation-error guarantees.