Multimodal neuroimaging data integration and pathway analysis.

Area of Science:

• Computational biology and bioinformatics
• Statistical genetics and genomics
• Neuroscience and neuroimaging

Background:

• Multimodal data collection (e.g., neuroimaging, multi-omics) is increasing in scientific research.
• Integrating diverse data types enables new mechanistic insights but lacks advanced analytical methodologies.
• Current methods struggle with the joint analysis of high-dimensional multimodal measurements.

Purpose of the Study:

• To develop a statistical framework for high-dimensional multimodal data integration.
• To investigate the mediating roles of different data modalities in exposure-outcome pathways.
• To apply the method to understand brain structure and function in sex-based language processing.

Main Methods:

• Proposed a novel mediation model framework for two distinct sets of mediators.
• Developed a penalized optimization approach for parameter estimation in multimodal data.
• Conducted asymptotic analysis for theoretical properties and simulations for finite-sample performance.

Main Results:

• The proposed penalized optimization method provides a robust approach for parameter estimation.
• Theoretical analysis confirms the estimator's properties.
• Simulations demonstrate good finite-sample performance, validating the method's practical utility.

Conclusions:

• The developed methodology effectively integrates high-dimensional multimodal data for mediation analysis.
• It advances the understanding of complex biological pathways by elucidating mediating roles of different data types.
• The brain pathway analysis illustrates the method's capability in real-world scientific applications.