A Bayesian Integrative Mixed Modeling Framework for Analysis of the Multi-Site Adolescent Brain and Cognitive...

Area of Science:

• Neuroscience
• Biostatistics
• Data Science

Background:

• Analyzing high-dimensional, heterogeneous data from multi-site cohort studies presents challenges in variable selection and prediction.
• Complex hierarchical structures in data require specialized analytical frameworks.

Purpose of the Study:

• To extend the Bayesian Integrative Analysis and Prediction (BIP) framework for simultaneous variable selection and outcome modeling.
• To develop a novel Bayesian Integrative Mixed Modeling (BIPmixed) framework for multi-view nested hierarchical data.
• To apply BIPmixed to the Adolescent Brain Cognitive Development (ABCD) Study for behavioral outcome prediction.

Main Methods:

• Developed the Bayesian Integrative Mixed Modeling (BIPmixed) framework.
• Incorporated 2-level nested random effects for enhanced interpretability and prediction in hierarchical data.
• Applied BIPmixed to multi-view data (structural/functional MRI, early life adversity) from the ABCD Study.

Main Results:

• BIPmixed effectively integrates multi-view data while accounting for nested sampling structures.
• Simulation studies demonstrated the robustness of BIPmixed across different random effect settings.
• Identified relevant variables and predicted behavioral outcomes in the ABCD Study.

Conclusions:

• BIPmixed is a valuable tool for analyzing large-scale studies with complex, hierarchical, and multi-view data.
• The framework enhances variable selection and prediction accuracy in nested data settings.
• Facilitates robust analysis of complex cohort studies like the ABCD Study.