Bayesian Inference of Multiple Gaussian Graphical Models.

Area of Science:

• Statistics
• Computational Biology
• Network Science

Background:

• Inferring network structures from data is crucial in many scientific fields.
• Analyzing multiple related networks presents challenges in identifying shared and unique features.
• Existing methods may not effectively leverage shared information across datasets.

Purpose of the Study:

• To develop a Bayesian approach for inferring multiple Gaussian graphical models.
• To identify and quantify shared network structures among related datasets.
• To incorporate prior knowledge and measure network relatedness.

Main Methods:

• Utilized a Markov random field (MRF) prior to link graph structure estimations and encourage common edges.
• Employed a spike-and-slab prior on network relatedness parameters to learn shared structures.
• Incorporated edge-specific informative priors for prior knowledge integration.

Main Results:

• Demonstrated improved accuracy in network estimation compared to related methods, particularly for moderate subgroup sample sizes.
• Successfully summarized relative network similarity across different sample groups.
• Validated the method's utility through simulations and application to protein networks in cancer subtypes.

Conclusions:

• The proposed Bayesian method effectively infers multiple Gaussian graphical models with shared structures.
• The approach provides a robust measure of network similarity and enhances estimation accuracy.
• Applicable to biological network inference, such as protein-protein interactions in cancer research.