Identifying diagnosis-specific genotype-phenotype associations via joint multitask sparse canonical correlation...

Area of Science:

• Neuroscience
• Genetics
• Biostatistics

Background:

• Brain imaging genetics investigates genotype-phenotype associations.
• Neurodegenerative disorders are heterogeneous, complicating analysis.
• Existing unsupervised methods fail to capture diagnosis-specific patterns.

Purpose of the Study:

• To develop a method for identifying diagnosis-specific genotype-phenotype associations.
• To address limitations of unsupervised sparse canonical correlation analysis (SCCA).
• To improve understanding of brain disorder pathophysiology.

Main Methods:

• Proposed a joint multitask learning framework, MT-SCCALR.
• Integrated SCCA with logistic regression for enhanced analysis.
• Developed an efficient optimization algorithm with guaranteed convergence.

Main Results:

• MT-SCCALR identifies diagnosis-specific and shared genotype-phenotype patterns.
• Achieved superior or comparable canonical correlation coefficients and classification performance.
• Demonstrated more discriminative canonical weight patterns than existing methods.

Conclusions:

• MT-SCCALR effectively identifies heterogeneous genotype-phenotype patterns in brain disorders.
• The method enhances understanding of the pathophysiology of brain disorders.
• MT-SCCALR offers a powerful tool for brain imaging genetics research.