A Bayesian model for classifying all differentially expressed proteins simultaneously in 2D PAGE gels.

Area of Science:

• Proteomics
• Bioinformatics
• Statistical Modeling

Background:

• Two-dimensional polyacrylamide gel electrophoresis (2D PAGE) is a standard technique for identifying differentially expressed proteins.
• Previous univariate probabilistic models (Wu et al., 2009) used a Likelihood Ratio Test (LRT) for differential expression analysis.
• Existing models account for missing data due to non-expression or levels below detection limits.

Purpose of the Study:

• To develop a global Bayesian model extending the univariate approach for analyzing differential protein expression.
• To simultaneously model all differentially expressed proteins, improving accuracy over individual protein analysis.

Main Methods:

• Developed a global Bayesian model incorporating global distributions to model relationships between protein parameters.
• Utilized Markov chain Monte Carlo (MCMC) integration for parameter recovery.
• Employed 95% highest posterior density (HPD) intervals to assess differences in mean expression intensities and probabilities of expression.

Main Results:

• The global Bayesian model accurately recovers underlying distributions and identifies more differentially expressed proteins compared to a simple LRT.
• Simulation analyses confirm the model's effectiveness in simultaneous differential expression analysis.
• The developed model demonstrates a very low False Discovery Rate (FDR).

Conclusions:

• The global Bayesian model offers a more powerful approach for differential protein expression analysis in 2D PAGE.
• This method provides more accurate parameter estimates by leveraging information across all proteins.
• The source code for the BIDE-2D model is publicly available.