Apparently low reproducibility of true differential expression discoveries in microarray studies

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Gene expression analysis from microarray studies often yields inconsistent results for the same disease.
• Low reproducibility is observed even in technical replicates, leading to concerns about false discoveries in small studies.

Purpose of the Study:

• To investigate the causes of low reproducibility in differentially expressed gene (DEG) detection.
• To evaluate the reliability of DEG lists from microarray data.
• To propose new methods for assessing reproducibility in complex disease studies.

Main Methods:

• Development and application of a statistical model to analyze DEG detection reproducibility.
• Analysis of simulated and real cancer gene expression data.
• Assessment of false discovery rates (FDR) in DEG lists.

Main Results:

• Small measurement variations in technical replicates cause significant inconsistency in DEG lists.
• Heterogeneous biological variations in cancer data further decrease reproducibility.
• Despite low reproducibility, individual DEG lists often have a low FDR, indicating a high proportion of true discoveries.
• Existing methods for comparing DEG lists are insufficient for complex diseases.

Conclusions:

• Low reproducibility in DEG detection does not necessarily imply low microarray technology quality.
• Biological variability significantly impacts DEG detection reproducibility.
• Novel metrics are needed to evaluate the reproducibility of molecular discoveries in complex diseases, considering correlated changes.