XDE

XDE estimates differential gene expression across multiple studies and microarray platforms using a hierarchical Bayesian model to account for data heterogeneity.

Key Features:

• Hierarchical Bayesian Framework: Implements a hierarchical Bayesian model that integrates expression data from multiple studies to accommodate between-study variability.
• Shrinkage Across Genes and Studies: Applies shrinkage across genes and studies to stabilize differential expression estimates and mitigate overfitting.
• Flexible Modeling for Platform Interactions: Models interactions between platforms (e.g., cDNA, Affymetrix) and estimated gene expression effects, allowing concordant and discordant differential expression across studies.
• Performance Evaluation: Evaluates performance using artificial data simulations and a split-study validation approach under both null and alternative hypotheses.
• Advantages in Small Studies: Demonstrates improved performance over direct combinations of t- and SAM-statistics in small studies when assessed by AUC, FDR, and MDR.
• Appropriate Shrinkage: Exhibits appropriate shrinkage in scenarios without platform-, sample-, or annotation-differences.

Scientific Applications:

• Breast cancer differential expression (ER+ vs ER−): Applied to four breast cancer microarray studies using different technologies to estimate differential expression between estrogen receptor positive and negative tumors.

Methodology:

Uses a hierarchical Bayesian model that integrates data from multiple sources, applies shrinkage across genes and studies, and flexibly models platform-by-effect interactions; performance assessed via artificial data simulations and split-study validation.