Comparison of statistical methods and the use of quality control samples for batch effect correction in human...

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Population-based studies often require gene expression analysis across different batches, introducing technical variation.
• Batch effects can confound biological signals, necessitating robust correction methods.
• Evaluating methods for batch effect removal is crucial for accurate interpretation of large-scale genomic data.

Purpose of the Study:

• To compare the performance of linear mixed models (LMM) and Combat for batch effect removal in gene expression data.
• To assess the utility of incorporating quality control samples as technical replicates in study design.
• To evaluate these methods across various simulation parameters including effect sizes, noise levels, and design balance.

Main Methods:

• Simulated gene expression data by introducing 'treatment' and batch effects to a real dataset.
• Applied linear mixed models (LMM) and Combat for batch effect correction.
• Assessed performance using sensitivity and specificity under diverse simulation conditions.

Main Results:

• Linear mixed models (LMM) detected stronger associations for large effect sizes compared to Combat.
• Combat identified a higher number of true and false positives than LMM.
• Quality control samples did not significantly reduce batch effects when using either LMM or Combat.

Conclusions:

• Both LMM and Combat are effective for batch effect removal, with subtle differences in their performance characteristics.
• The choice between LMM and Combat may depend on specific research objectives, particularly regarding the balance of true/false positives.
• In this study, quality control samples provided no added value for mitigating batch effects in the analyzed gene expression data.