FDR-Corrected Sparse Canonical Correlation Analysis With Applications to Imaging Genomics

Area of Science:

• Neuroimaging
• Genomics
• Statistical analysis

Background:

• High-dimensional data in life sciences, particularly neuroimaging and genomics, presents challenges in reducing false discoveries.
• The false discovery rate (FDR) is a crucial metric for hypothesis testing in these fields.
• Sparse Canonical Correlation Analysis (CCA) is used for high-dimensional cross-correlation analysis.

Purpose of the Study:

• To propose and validate a method for applying and controlling the false discovery rate (FDR) within sparse CCA.
• To adapt the sparsity of the solution to the true underlying sparsity level in the data.

Main Methods:

• Developed a novel FDR control procedure specifically for sparse CCA.
• The method directly influences solution sparsity, adapting it to unknown sparsity levels.
• Validated through theoretical derivations and simulation studies.

Main Results:

• The proposed FDR correction method effectively controls the FDR of canonical vectors below a user-defined threshold.
• Demonstrated the method's efficacy in high-dimensional settings.
• Successfully applied to an imaging genomics dataset from the Philadelphia Neurodevelopmental Cohort.

Conclusions:

• The developed FDR control method is effective for sparse CCA in high-dimensional data.
• The approach successfully links brain connectivity patterns to genomic data.
• This method offers a robust tool for analyzing complex neuroimaging and genomics datasets.