RNA sequencing (RNA-seq) has become a cornerstone of transcriptomics, providing detailed insights into gene expression across diverse biological conditions and sample types. However, RNA-seq data are often confounded by batch effects, systematic non-biological variations that compromise data reliability and obscure true biological differences. To address these challenges, we introduce ComBat-ref, a refined batch effect correction method designed to enhance the statistical power and reliability of differential expression analysis in RNA-seq data. Building on the principles of ComBat-seq, ComBat-ref employs a negative binomial model for count data adjustment but innovates by selecting a reference batch with the smallest dispersion, preserving count data for the reference batch, and adjusting other batches towards the reference batch. Our method demonstrated superior performance in both simulated environments and real-world datasets, including the growth factor receptor network (GFRN) data and NASA GeneLab transcriptomic datasets, significantly improving sensitivity and specificity compared to existing methods. By effectively mitigating batch effects while maintaining high detection power, ComBat-ref provides a robust solution for improving the accuracy and interpretability of RNA-seq data analyses.
Keywords: Batch effect correction; Generalized linear model; Minimum dispersion; Negative binomial distribution; RNA-seq data.
© 2024 The Authors.