HEM

HEM models gene-level and condition-specific expression variability in genome-wide microarray data using a Bayesian hierarchical framework to account for heterogeneous experimental and biological error components.

Key Features:

• Bayesian hierarchical framework: Implements a hierarchical error model that decomposes variability across genes and conditions.
• Error decomposition: Separates experimental and biological error components when both types of replicates are available.
• Markov chain Monte Carlo (MCMC) inference: Uses MCMC to estimate a large number of parameters from a unified likelihood function across all genes.
• F-like summary statistic: Computes an F-like statistic from HEM estimates to facilitate identification of differentially expressed genes under multiple conditions.
• Heterogeneous error handling: Models heterogeneous error variability across intensity ranges and between genes.
• Gene-specific parameter estimation: Estimates gene-level genetic parameters and interaction expression patterns across multiple biological conditions.
• Comparative evaluation: Performance assessed via simulations and comparisons with ANOVA and published microarray datasets.

Scientific Applications:

• Genome-wide microarray analysis: Analysis of genome-wide microarray data with heterogeneous error structures.
• Differential expression detection: Identification of differentially expressed genes across multiple conditions using an F-like summary statistic.
• Gene- and interaction-level inference: Estimation of gene-specific parameters and interaction expression patterns across biological conditions.
• Method validation and benchmarking: Validation and benchmarking via simulation studies and analysis of primate brain and mouse B-cell development microarray datasets.

Methodology:

Employs a Bayesian hierarchical framework that decomposes error into experimental and biological components (requiring respective replicates); performs inference by Markov chain Monte Carlo on a unified likelihood across genes; derives an F-like summary statistic from HEM estimates; validation via simulations and comparison with ANOVA and application to primate brain and mouse B-cell development microarray datasets.