Generalized Bayesian Factor Analysis for Integrative Clustering with Applications to Multi-Omics Data.

Area of Science:

• Computational biology
• Bioinformatics
• Statistical genetics

Background:

• Integrative clustering analyzes multiple datasets for a common subject group, crucial for multi-omics data.
• Incorporating prior structural knowledge, like gene regulatory networks, into genomic studies is gaining interest.
• Existing methods may not fully leverage graph information for enhanced biological insights.

Purpose of the Study:

• To propose a novel integrative clustering method that incorporates prior graph knowledge.
• To extend the iCluster+ framework using a generalized Bayesian factor analysis (GBFA) approach.
• To improve the precision and biological interpretability of multi-omics data analysis.

Main Methods:

• Developed a generalized Bayesian factor analysis (GBFA) framework incorporating graph information.
• Employed spike and slab lasso (SSL) prior for sparse factor loadings and Markov random field (MRF) prior for smoothing.
• Proposed a variational Expectation-Maximization (EM) algorithm for efficient estimation of factor loadings.

Main Results:

• The GBFA framework successfully integrates graph structure and feature information.
• The proposed method demonstrated superior performance in simulation studies.
• Application to the NCI60 cell line dataset yielded more biologically meaningful outcomes compared to existing methods.

Conclusions:

• The novel integrative clustering method effectively incorporates prior graph knowledge.
• The GBFA framework provides a unified approach to shrinkage adaptive to loading size and graph structure.
• The method offers a powerful tool for identifying disease subtypes and understanding complex biological processes from multi-omics data.