Sparse semiparametric canonical correlation analysis for data of mixed types

Area of Science:

• Biostatistics
• Genomics
• Computational Biology

Background:

• Canonical correlation analysis (CCA) identifies relationships between variable sets.
• Traditional CCA struggles with high-dimensional and mixed data types (continuous, binary, zero-inflated).

Purpose of the Study:

• To develop a robust sparse canonical correlation analysis (SCCA) for high-dimensional, mixed-type data.
• To address limitations of existing CCA methods in modern biological datasets.

Main Methods:

• Proposed a semiparametric SCCA approach.
• Utilized truncated latent Gaussian copulas to model excess zeroes without marginal transformation estimation.
• Derived a rank-based estimator for the latent correlation matrix.

Main Results:

• The novel SCCA method demonstrates effectiveness in high-dimensional settings.
• Successfully applied to analyze associations between gene expression and microRNA data in breast cancer patients.

Conclusions:

• The proposed semiparametric SCCA offers a powerful tool for analyzing complex biological data.
• This approach advances the study of relationships in high-dimensional, mixed-type datasets, particularly in cancer genomics.