Multi-Task Sparse Canonical Correlation Analysis with Application to Multi-Modal Brain Imaging Genetics.

Area of Science:

• Neuroscience
• Genetics
• Computational Biology

Background:

• Brain imaging genetics integrates genetic data (SNPs) with brain imaging quantitative traits (QTs) to understand brain structure and function.
• Existing methods like Multi-Task Learning (MTL) and Sparse Canonical Correlation Analysis (SCCA) have limitations in handling multiple QTs and computational efficiency with increasing SNPs.
• MTL methods struggle with feature selection across multiple QTs, while SCCA typically uses one QT modality, and both can be computationally intensive.

Purpose of the Study:

• To propose a novel Multi-Task Sparse Canonical Correlation Analysis (MT-SCCA) method for identifying bi-multivariate associations between SNPs and multi-modal imaging QTs.
• To leverage complementary information from different imaging modalities.
• To develop an efficient and scalable method for genome-wide and brain-wide imaging genetics.

Main Methods:

• Developed MT-SCCA, enforcing group-level sparsity using the G2,1-norm for QTs.
• Employed the l2,1-norm for joint feature selection across multiple tasks (SNPs and QTs).
• Proposed a fast optimization algorithm incorporating SNP grouping information.

Main Results:

• MT-SCCA demonstrated superior correlation coefficients and canonical weight patterns compared to conventional SCCA.
• The proposed method showed significant improvements in computational speed and ease of implementation.
• MT-SCCA effectively utilizes complementary information from multi-modal imaging data.

Conclusions:

• MT-SCCA offers a powerful and efficient approach for brain imaging genetics studies.
• The method facilitates the identification of complex genetic associations with brain structure and function.
• MT-SCCA holds significant potential for advancing genome-wide and brain-wide genetic analyses.