Network-based analysis of multivariate gene expression data

Area of Science:

• Genomics
• Bioinformatics
• Systems Biology

Background:

• Multivariate gene expression data are crucial for understanding genomic responses under ordered conditions.
• Identifying genes with differential expression patterns over time or dosage is key to uncovering biological processes.
• Existing empirical Bayes methods often assume gene independence, limiting their effectiveness.

Purpose of the Study:

• To develop an alternative empirical Bayes approach for multivariate gene expression data analysis.
• To model gene dependencies using a discrete Markov random field (MRF) prior.
• To improve the identification of differentially expressed genes and perturbed pathways.

Main Methods:

• Introduced a novel empirical Bayes approach incorporating a discrete Markov random field (MRF) prior.
• Modeled dependencies between differential expression patterns of genes within biological networks.
• Utilized simulation studies and a real-world microarray time course dataset.

Main Results:

• The proposed MRF-based method effectively identifies differentially expressed genes and subnetworks.
• Demonstrated higher sensitivity compared to methods that ignore pathway information.
• Achieved similar false discovery rates to existing procedures.

Conclusions:

• The MRF prior approach offers a powerful tool for analyzing multivariate gene expression data by leveraging gene network dependencies.
• The method successfully identified key genes and subnetworks in MAPK, focal adhesion, and prion disease pathways.
• These findings provide insights into cell differentiation mechanisms in neuroblastoma cell lines.