NExON-Bayes: A Bayesian approach to network estimation informed by ordinal covariates.

Area of Science:

• Computational Biology
• Bioinformatics
• Statistical Genetics

Background:

• Omic network analysis is vital for understanding complex diseases.
• Existing methods often assume data homogeneity, leading to inaccurate network estimates.
• Accounting for sample variability, such as disease stage, is crucial for reliable omic network analysis.

Purpose of the Study:

• To develop a joint Gaussian graphical model that accounts for sample-level ordinal covariates to improve omic network estimation.
• To address the challenge of heterogeneity in disease settings for more interpretable omic network estimates.
• To provide a robust framework for analyzing dynamic biological networks across disease progression.

Main Methods:

• Proposed NExON-Bayes, an extension of the graphical spike-and-slab framework incorporating ordinal covariates.
• Developed an efficient variational inference algorithm for high-dimensional omic data.
• Validated the method through simulations and application to breast carcinoma proteomic data.

Main Results:

• NExON-Bayes demonstrated superior performance compared to existing network approaches in simulations.
• The method accurately estimated proteomic network structures and covariate dependencies in breast cancer.
• Identified shifts in biological pathways across different stages of breast carcinoma progression.

Conclusions:

• NExON-Bayes effectively models heterogeneity in omic data using ordinal covariates.
• The framework provides comprehensive insights into dynamic biological networks and disease progression.
• The developed R package facilitates the application of this advanced network analysis technique.