Adjusting for Batch Effects in DNA Methylation Microarray Data, a Lesson Learned.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Molecular data, including low- and high-throughput data, can contain systematic technical variation known as batch effects.
• Batch effects can be confounded with variables of interest, leading to misinterpretation of technical variation as biological findings.
• Identifying and correcting batch effects is essential for reducing false discoveries in molecular data analysis.

Purpose of the Study:

• To demonstrate the impact of batch effects in DNA methylation data analysis.
• To evaluate the effectiveness of the ComBat R function in adjusting for batch effects.
• To highlight the importance of study design and data processing in accurate molecular data analysis.

Main Methods:

• Analysis of a 30-sample pilot Illumina Infinium HumanMethylation450 (450k array) experiment.
• Application of the ComBat R function to adjust for identified row and chip batch effects.
• Comparison of an "initial analysis" with an unbalanced design against a "revised processing" with improved design and more samples.

Main Results:

• The initial analysis using ComBat on an unbalanced design introduced thousands of false positive DNA methylation differences.
• Revised processing, incorporating study design changes and more samples, successfully reduced batch effects without introducing false signals.
• Batch effects, specifically row and chip effects, were identified in the 450k array experiment.

Conclusions:

• Thoughtful study design is the primary method to mitigate batch effects.
• All DNA methylation microarray analyses must include inspection, assessment, and correction of batch effects if present.
• Clear communication between technicians and analysts, and establishing research questions a priori, are vital for robust molecular data analysis.