Bayesian combinatorial MultiStudy factor analysis.

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Mutations in BRCA1 and BRCA2 genes are strongly linked to breast cancer risk.
• Understanding shared and unique transcript expression patterns in mutation carriers is crucial for differentiating disease mechanisms.
• Existing methods like Bayesian Multi-Study Factor Analysis (BMSFA) have limitations in identifying partially shared signals.

Purpose of the Study:

• To introduce Tetris, a novel Bayesian combinatorial multi-study factor analysis method.
• To extend BMSFA by enabling the identification of latent factors shared across any combination of studies or groups.
• To apply Tetris to transcript expression data from high-risk families to characterize mutation-specific features and pathways.

Main Methods:

• Developed Tetris, a Bayesian method using the Indian Buffet Process to model shared latent factors across subsets of studies.
• Incorporated credible balls for summarizing uncertainty in factor sharing patterns.
• Validated Tetris through extensive simulations for dimension reduction and covariance estimation.

Main Results:

• Tetris successfully identified latent factors shared by specific combinations of BRCA1 and BRCA2 mutation groups.
• The method revealed distinct transcriptomic features and pathways characterizing each mutation group and their interrelationships.
• An extension of Tetris demonstrated utility in discovering sample groupings without pre-defined labels.

Conclusions:

• Tetris provides a powerful framework for analyzing complex, high-dimensional multi-study data, particularly in genomics.
• The method offers deeper insights into the heterogeneity of breast cancer based on BRCA mutation status.
• Tetris facilitates the discovery of novel biological insights and potential therapeutic targets by uncovering shared and unique molecular signatures.