A Tutorial on Bayesian Multi-Study Factor Analysis With Applications in Nutrition and Genomics

Area of Science:

• Biomedical research
• Statistical modeling
• Genomics
• Nutrition science

Background:

• High-dimensional data integration is vital in biomedical research for enhanced statistical power and reproducibility.
• Factor analysis (FA) is a key dimensionality reduction technique, with Bayesian extensions offering powerful multi-study integration capabilities.
• Disentangling shared biological signals from study-specific variability is crucial for accurate data interpretation.

Purpose of the Study:

• To provide a practical, comparative guide to seven advanced Bayesian integrative factor models.
• To contextualize these methods with benchmark approaches like Standard FA (Stack FA) and individual FA (Ind FA).
• To offer a comprehensive analytical workflow with R code for real-world applications.

Main Methods:

• Comparative evaluation of seven Bayesian integrative factor models: PFA, MOM-SS, SUFA, BMSFA, CAVI, BLAST, and Tetris.
• Benchmarking against Standard FA on pooled data (Stack FA) and individual study FA (Ind FA).
• Extensive simulations to assess computational efficiency, accuracy of loading estimation, and factor number determination.

Main Results:

• The study evaluates the performance of various Bayesian integrative factor analysis models.
• Simulation results provide insights into the computational efficiency and accuracy of each method.
• The R package 'bmfaToolkits' and accompanying code facilitate practical implementation.

Conclusions:

• Bayesian integrative factor analysis provides robust tools for extracting interpretable patterns from complex, multi-source biomedical data.
• The tutorial equips applied researchers with practical guidance and computational resources for advanced data integration.
• This work enhances the ability to derive meaningful scientific insights from high-dimensional, heterogeneous datasets.