Comparison of two different stochastic models for extracting protein regulatory pathways using Bayesian networks

Area of Science:

• Toxicoproteomics
• Systems Biology
• Computational Biology

Background:

• Toxicoproteomics integrates toxicology and systems biology to understand biological responses to chemical exposures.
• Inferring biochemical regulatory networks is crucial for analyzing these responses.
• Previous work compared relevance networks, graphical Gaussian models, and Bayesian networks (BNs).

Purpose of the Study:

• To evaluate and compare the learning performance of two stochastic Bayesian network models: BGe and BDe.
• To assess these models using cytometric protein expression data from the RAF-signaling pathway.
• To determine the preferable model for inferring biological pathways.

Main Methods:

• Cross-comparison of Bayesian network (BN) learning algorithms (BGe and BDe).
• Utilized real cytometric protein expression data from the RAF-signaling pathway.
• Evaluated model performance based on inferred network structures.

Main Results:

• The study compared the learning performances of BGe and BDe Bayesian network models.
• The more flexible BDe model requires data discretization, potentially causing information loss.
• The BDe model demonstrated preferable performance over the BGe model with sufficient observational data.

Conclusions:

• The BDe Bayesian network model is recommended over the BGe model for toxicoproteomic pathway analysis when ample data is available.
• This finding aids in selecting appropriate computational methods for inferring biological regulatory networks.
• Understanding the RAF-signaling pathway is critical due to its link to carcinogenesis.