Bayesian linear mixed model with multiple random effects for prediction analysis on high-dimensional multi-omics...

Area of Science:

• Genomics
• Biostatistics
• Precision Medicine

Background:

• High-dimensional multi-omics data offer valuable resources for disease risk prediction.
• Analyzing complex inter/intra-relationships within multi-omics data presents significant analytical challenges.

Purpose of the Study:

• To introduce a novel statistical framework for accurate disease risk prediction using multi-omics data.
• To address the analytical challenges posed by high-dimensional and complex multi-omics data.

Main Methods:

• Proposed a two-step Bayesian linear mixed model framework (TBLMM).
• Employed a hybrid of sparsity regression and linear mixed models with multiple random effects.
• Utilized kernel fusion to model non-linear and interaction effects within multi-omics data.
• Implemented a computationally efficient variational Bayes algorithm for parameter inference.

Main Results:

• TBLMM effectively models predictive effects from multi-omics data, capturing complex relationships.
• The framework accommodates non-linear and interaction effects through kernel fusion.
• Extensive simulations and real-world data analysis (Alzheimer's Disease Neuroimaging Initiative) demonstrated TBLMM's superior performance.
• TBLMM consistently outperformed existing methods in predicting complex trait risks.

Conclusions:

• TBLMM provides a robust and effective framework for disease risk prediction using multi-omics data.
• The method's ability to handle complex data structures offers advancements in precision medicine.
• An R package for TBLMM is publicly available on GitHub for broader application.