Robust Bayesian graphical regression models for assessing tumor heterogeneity in proteomic networks.

Area of Science:

• Computational Biology
• Statistical Modeling
• Bioinformatics

Background:

• Graphical models are essential for analyzing high-throughput biological data.
• Existing models often assume data normality and graph homogeneity, limiting their application in complex biological systems like cancer proteomic networks.

Purpose of the Study:

• To develop a novel statistical framework, robust Bayesian graphical regression (rBGR), for estimating heterogeneous graphs from non-normally distributed data.
• To address the limitations of existing graphical models in capturing complex dependency structures in biological data.

Main Methods:

• Proposed rBGR, a flexible framework accommodating non-normality via random marginal transformations.
• Incorporated covariate-dependent graphs using graphical regression techniques.
• Introduced conditional sign independence with covariates and an efficient posterior sampling algorithm.

Main Results:

• rBGR demonstrated superior performance over existing models in simulation studies for both edge and covariate selection, especially under non-normal data conditions.
• Applied rBGR to analyze proteomic networks in lung and ovarian cancers, investigating immunogenic heterogeneity.
• Identified significant protein-protein interactions differentially associated with immune cell abundance, including novel findings.

Conclusions:

• rBGR provides a robust and flexible approach for modeling heterogeneous and non-normal biological network data.
• The method offers valuable insights into cancer proteomic networks and their relationship with tumor immunogenicity.
• rBGR facilitates the discovery of biologically relevant protein interactions, advancing cancer research.