Bayesian state-space modeling in gene expression data analysis: An application with biomarker prediction.

Area of Science:

• Biostatistics
• Bioinformatics
• Genomics

Background:

• Bayesian State Space models offer advanced stochastic modeling.
• These models integrate prior knowledge and observed data likelihood to capture hidden process randomness.
• Their application in analyzing longitudinal gene expression data is explored.

Purpose of the Study:

• To elucidate the scope of Bayesian state space modeling for predicting future expression values in longitudinal microarray data.
• To demonstrate the flexibility of Bayesian state space models with different covariance structures.

Main Methods:

• Utilized longitudinal clinical trial data (GSE30531) from NCBI GEO.
• Employed t-test for selecting differentially expressed genes.
• Applied Markov Chain Monte Carlo (MCMC) with Gibbs Sampling for parameter and future expression estimation.
• Assessed Variance Components, First order Auto Regressive, and Unstructured covariance structures.

Main Results:

• Selected 72 distinct genes with significant expression differences for model fitting.
• Parameter estimates showed consistent trends across various covariance structures.
• Cross-tabulation revealed a significant P-value (0.02) for gene frequencies based on credible intervals and study groups.

Conclusions:

• Bayesian state space models are effective for explaining and predicting complex gene expression data.
• The models demonstrate flexibility in handling different covariance structures.
• This approach provides a robust framework for analyzing high-dimensional biological data.