Removing Batch Effects from Longitudinal Gene Expression - Quantile Normalization Plus ComBat as Best Approach for...

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Technical variation, particularly batch effects, significantly impacts microarray gene expression data.
• Removing technical noise while preserving biological signals is crucial for reliable study findings.
• Existing batch effect removal methods lack thorough evaluation in large-scale longitudinal gene expression datasets.

Purpose of the Study:

• To identify an optimal method for batch effect removal in longitudinal microarray gene expression data.
• To evaluate the performance of various batch correction techniques on a large cohort.
• To ensure the preservation of biological variability after technical noise reduction.

Main Methods:

• Applied Deming regression, Passing-Bablok regression, linear mixed models, non-linear models, ReplicateRUV, and ComBat for batch effect correction.
• Utilized quantile normalization prior to batch effect correction for each method.
• Assessed technical variation using principal component analysis and evaluated biological variability through association analyses (e.g., BMI and transcriptomes).

Main Results:

• Quantile normalization combined with ComBat demonstrated successful reduction of batch effects.
• This approach effectively maintained biological variability, as evidenced by association analyses.
• ReplicateRUV performed well on replicates but failed on the full dataset; other methods showed limited batch effect reduction.

Conclusions:

• Quantile normalization followed by ComBat is a robust and valuable strategy for batch correction in longitudinal gene expression studies.
• This method balances the removal of technical variation with the preservation of biologically relevant gene expression patterns.
• The findings provide guidance for researchers analyzing large-scale longitudinal gene expression data.