Graph-guided Bayesian Factor Model for Integrative Analysis of Multi-modal Data with Noisy Network Information.

Area of Science:

• * Multimodal Data Analysis
• * Statistical Learning
• * Bioinformatics

Background:

• * Existing factor analysis methods struggle to integrate biological network knowledge into multimodal data analysis.
• * Graph-guided methods improve accuracy but often rely on incomplete or noisy external network data.
• * Functional genomics and metabolomics data require sophisticated approaches to capture complex biological structures.

Purpose of the Study:

• * To develop a novel graph-guided Bayesian factor model for multimodal data analysis.
• * To incorporate and account for noise within biological network information.
• * To identify globally shared, partially shared, and modality-specific latent factors.

Main Methods:

• * Proposed a Bayesian factor model incorporating network information from existing databases and estimated graphs.
• * Utilized a latent scale modeling framework to handle network noise.
• * Employed shrinkage priors for feature and modal-wise sparsity, enabling feature selection.
• * Developed an efficient Markov chain Monte Carlo algorithm for posterior sampling.

Main Results:

• * The proposed model successfully identifies different types of latent factors in multimodal data.
• * Demonstrated superior performance compared to existing methods through simulations.
• * Applied the model to gene expression and metabolomics data for Alzheimer's disease research.

Conclusions:

• * The graph-guided Bayesian factor model offers an interpretable and accurate approach for multimodal data analysis.
• * The method effectively integrates noisy biological network information for improved factor identification.
• * This approach has significant potential for applications in complex diseases like Alzheimer's using functional genomics and metabolomics data.