TOSCCA: a framework for interpretation and testing of sparse canonical correlations

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-dimensional omics and imaging data are routinely collected in biomedical research.
• Multivariate methods like Canonical Correlation Analysis (CCA) integrate datasets to uncover biological mechanisms.
• Interpretability of exploratory methods like CCA is crucial for biological insight.

Purpose of the Study:

• To present a novel sparse CCA method based on soft-thresholding.
• To improve the interpretability and computational efficiency of CCA for high-dimensional data integration.
• To offer a robust alternative to existing sparse CCA methods, avoiding burdensome parameter tuning.

Main Methods:

• Developed a sparse CCA method utilizing soft-thresholding for component generation.
• Implemented permutation-based hypothesis testing for statistical validation.
• Compared the proposed method against Penalized Matrix Analysis (PMA) using simulations and real cancer genomics data.

Main Results:

• The soft-thresholding approach yields near-orthogonal components and avoids penalty parameter tuning.
• The method demonstrates reduced dependency on initialization compared to alternatives.
• Real-world application on cancer genomics data showed improved interpretability and comparable or enhanced signal discovery versus PMA.

Conclusions:

• The proposed soft-thresholding sparse CCA method offers superior interpretability for high-dimensional data integration.
• This approach facilitates more effective discovery of underlying biological mechanisms from complex datasets.
• The method provides a computationally efficient and robust alternative for analyzing integrated omics and imaging data.