RFCCA

RFCCA estimates conditional canonical correlations between two sets of variables while accounting for subject-related covariates such as age, gender, or clinical measures.

Key Features:

• R library implementation: Implemented as an R library for estimation of conditional canonical correlations.
• Conditional canonical correlation estimation: Extends canonical correlation analysis (CCA) by incorporating subject-related covariates to estimate canonical correlations conditional on those covariates.
• Random forest framework: Employs a random forest approach that builds individual trees using a specialized splitting rule designed to maximize canonical correlation heterogeneity between child nodes.
• Significance testing: Provides a significance test to assess the global effect of covariates on the relationship between the two variable sets.
• Simulation validation: Performance and the significance test were evaluated via simulation studies showing accurate canonical correlation estimation with well-controlled Type-1 error rates.
• Practical application: Demonstrated on EEG data to illustrate extraction of conditional multivariate relationships in complex datasets.

Scientific Applications:

• Neuroscience (EEG analysis): Quantifies conditional relationships between neural measures and other variable sets in EEG studies.
• Clinical research: Assesses multivariate associations conditional on clinical covariates in clinical and biomedical studies.
• Multivariate data analysis: Applies to any domain requiring estimation of conditional relationships between two variable sets while accounting for subject-specific covariates.

Methodology:

Extends CCA by incorporating covariates, uses a random forest approach with a specialized splitting rule that maximizes canonical correlation heterogeneity between child nodes, includes a significance test for the global effect of covariates, and was validated through simulation studies.