Random rotation for identifying differentially expressed genes with linear models following batch effect correction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput technologies generate data susceptible to batch effects.
• Current methods often separate batch effect correction and hypothesis testing, potentially biasing results.
• Accurate statistical inference is critical for interpreting large-scale biological data.

Purpose of the Study:

• To develop a novel approach for estimating null distributions in data analysis pipelines with batch effect correction.
• To enable accurate calculation of P-values and false discovery rates following batch effect correction.
• To maintain the alpha level in statistical testing after batch effect correction.

Main Methods:

• The study proposes a method based on generating simulated datasets using random rotation.
• This approach effectively preserves the dependence structure among genes.
• It allows for the estimation of null distributions for dependent test statistics.

Main Results:

• The novel approach accurately estimates null distributions of test statistics.
• Resampling-based P-values and false discovery rates can be reliably calculated post-batch effect correction.
• The method maintains the desired alpha level, ensuring statistical validity.

Conclusions:

• The developed method provides a robust solution for statistical analysis of high-throughput data affected by batch effects.
• This approach enhances the reliability of P-values and false discovery rates.
• The randRotation package is available on Bioconductor for practical implementation.