BFRM

BFRM models sparse latent factors in large-scale gene expression datasets using Bayesian factor regression to infer pathway activity and support predictive molecular profiling.

Key Features:

• Sparse Latent Factor Models: Employs sparse latent factor models to decompose gene expression data into interpretable components representing underlying biological pathways.
• Pathway Analysis: Identifies and characterizes heterogeneity within oncogenic pathways and links factor structure to clinical biomarkers.
• Hierarchical Sparsity Priors: Uses hierarchical sparsity priors to perform dimension reduction and address multiple comparisons in high-dimensional data.
• Non-Gaussian/Nonparametric Components: Incorporates non-Gaussian and nonparametric model components to capture complex multivariate expression patterns.

Scientific Applications:

• Cancer Pathway Analysis: Applied to breast cancer studies to investigate the structure and activity of oncogenic pathways and their relation to clinical outcomes.
• Molecular Profiling and Predictive Modeling: Enables molecular profiling and development of predictive models from large-scale gene expression data.
• Biological Activity Interpretation: Supports overlaying statistical factors onto known biological activities to interpret disease mechanisms and biomarker associations.

Methodology:

Applies sparsity modeling techniques to multivariate regression, ANOVA, and latent factor models, employs hierarchical sparsity priors and non-Gaussian/nonparametric components, and performs model search and fitting using stochastic simulation and evolutionary stochastic search methods.