Bayesian learning of multiple directed networks from observational data.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Graphical modeling is crucial for understanding biological networks like gene regulatory and protein interaction networks.
• Biological data often shows heterogeneity across groups (e.g., disease subtypes), affecting network structures and pathways.
• Existing methods may not fully capture both shared similarities and unique differences across these groups.

Purpose of the Study:

• To develop a novel Bayesian method for the joint structural learning of multiple directed acyclic graphs (DAGs).
• To effectively model heterogeneity in biological data while leveraging similarities across different groups.
• To provide a robust framework for analyzing complex biological networks with group-specific variations.

Main Methods:

• Developed a Bayesian approach for structural learning of multiple DAGs, explicitly considering Markov equivalence.
• Introduced a prior on graph spaces that enables selective strength borrowing across groups.
• Utilized posterior probability of edge inclusion for inferring network structures and flow directions.

Main Results:

• The proposed method successfully infers multiple DAGs by accommodating both shared patterns and group-specific idiosyncrasies.
• Simulation studies demonstrated the comparative performance of the developed method.
• Analysis of two protein networks provided substantive interpretations of biological findings.

Conclusions:

• The Bayesian framework offers a powerful tool for joint inference of multiple biological networks with heterogeneous data.
• The method enhances the understanding of complex dependencies and pathway modifications in distinct biological groups.
• This approach facilitates more accurate and interpretable network analysis in systems biology and bioinformatics.