ENFORCING CO-EXPRESSION IN MULTIMODAL REGRESSION FRAMEWORK.

Area of Science:

• Neuroscience
• Genetics
• Statistical Modeling

Background:

• Complex neurological diseases like schizophrenia involve intricate genetic and neurological factors.
• Integrating multimodal data (e.g., genetic and neuroimaging) is crucial for understanding these diseases.
• Existing methods like Lasso regression and Canonical Correlation Analysis (CCA) have limitations in multimodal integration.

Purpose of the Study:

• To develop a novel exploratory multivariate method for multimodal data integration.
• To estimate the link between genetic variability and neurological variability in disease studies.
• To extract discriminative variables co-expressed across different data modalities.

Main Methods:

• Proposed a new method combining Lasso regression and Canonical Correlation Analysis (CCA).
• Utilized a 'CCA-type' formulation to regularize multimodal Lasso regression.
• Evaluated the method on simulated data and real-world SNP and fMRI data.

Main Results:

• The developed method successfully integrates genetic and neuroimaging data.
• Identified discriminative variables that show co-expression across modalities.
• Demonstrated the method's utility in the context of schizophrenia research.

Conclusions:

• The proposed integrated method offers a promising approach for multimodal data analysis in neuroscience.
• This technique can enhance the understanding of complex neurological diseases by linking genetic and functional brain data.
• Further validation and application in diverse disease contexts are warranted.