Distance canonical correlation analysis with application to an imaging-genetic study.

Area of Science:

• Neuroscience
• Genetics
• Biostatistics

Background:

• Distance correlation detects linear and nonlinear associations but requires multiple testing correction in imaging genetic studies, reducing sensitivity.
• Large-scale inferences in imaging genetics necessitate robust statistical methods for reliable association detection.

Purpose of the Study:

• To develop a novel method, distance canonical correlation analysis (DCCA), to enhance the detection of complex associations in multimodal imaging-genetic data.
• To overcome the sensitivity limitations of traditional distance correlation in high-dimensional imaging genetic analyses.

Main Methods:

• DCCA constructs a distance kernel function and solves an optimization problem to identify feature combinations with maximal distance correlation.
• The method integrates distance correlation with canonical correlation analysis principles for enhanced feature selection.

Main Results:

• DCCA successfully identified mental disorder-related gene pathways and brain networks in the Philadelphia Neurodevelopmental Cohort (PNC) dataset.
• Analysis revealed nonlinear connections within the default mode network and other brain networks.
• Brain network connectivity showed developmental trajectories, strengthening with age, particularly during adolescence.

Conclusions:

• DCCA offers a powerful approach for analyzing complex multimodal brain imaging and genetic data.
• The method enhances the detection of both linear and nonlinear associations, proving valuable for neurodevelopmental studies.
• Findings highlight the importance of adolescence in brain network development and identify potential genetic underpinnings of mental disorders.