Belief propagation in genotype-phenotype networks.

Area of Science:

• Genomics and Systems Biology
• Computational Biology and Bioinformatics
• Network Science

Background:

• Graphical models are crucial for linking genotypes to phenotypes.
• Network inference methods leverage natural genetic variation.
• Current analyses often end with network structure, limiting deeper insights.

Purpose of the Study:

• To extend graphical analyses by exploring network perturbations and quantifying system-wide effects.
• To develop a method for perturbing phenotype-genotype networks and assessing node-wise impacts.
• To provide a framework for hypothesis generation beyond static network structures.

Main Methods:

• Utilized belief propagation in Conditional Gaussian Bayesian Networks (CG-BNs) for evidence propagation.
• Modeled genotype-phenotype networks as a specific subclass of CG-BNs ensuring exact inference.
• Quantified system-wide effects using symmetric Kullback-Leibler divergence between perturbed and unperturbed marginal distributions.
• Developed the geneNetBP software package in R for implementing the approach.

Main Results:

• Successfully predicted and visualized system-wide network effects in response to phenotypic perturbations.
• Identified coordinated responses in sub-pathways and network regions, suggesting co-regulation.
• Demonstrated the stability of predictions across an ensemble of likely network structures.
• Applied the method to kidney and skin cancer expression quantitative trait loci (eQTL) data in Mus musculus.

Conclusions:

• The developed method effectively quantifies system-wide network effects from phenotypic changes.
• Findings highlight co-regulation and coordination within genetic networks in response to altered phenotypes.
• Predictions can be examined alongside covariates like cancer status, offering biological insights.
• The approach provides a robust tool for exploring dynamic network behavior in biological systems.