Probabilistic edge inference of gene networks with markov random field-based bayesian learning

Area of Science:

• Computational Biology
• Network Science
• Genomics

Background:

• Gene regulatory network (GRN) construction often relies on Gaussian graphical models (GGMs).
• Existing GGM algorithms typically make deterministic edge presence/absence decisions.
• Probabilistic inference of edge existence and strength is frequently overlooked due to computational or implementation challenges.

Purpose of the Study:

• To develop a method that simultaneously infers probabilistic edge existence and quantifies edge strength in GRNs.
• To address the limitations of deterministic edge decisions in current GRN construction algorithms.
• To provide a framework for prioritizing edges based on their quantified strength for biological interpretation.

Main Methods:

• Integration of Bayesian Markov random field and conditional autoregressive (CAR) models.
• Application of spike-and-slab lasso prior for quantifying uncertainty and strength.
• Utilizing edge strength for prioritizing network components.

Main Results:

• The proposed Bayesian CAR model effectively quantifies both edge uncertainty and relative strength.
• Simulations and a glioblastoma cancer study demonstrated stable performance.
• The approach identified novel network structures, offering potential biological insights.

Conclusions:

• The combined Bayesian Markov random field and CAR model offers a robust framework for GRN construction.
• This probabilistic approach enhances biological interpretability by quantifying edge importance.
• The method shows promise for uncovering complex gene regulatory relationships and prioritizing key interactions.