Evaluation of statistical differential analysis methods for identification of senescent cells using single-cell transcriptomics

Differential gene expression (DGE) analysis is a crucial step in identifying senescent cells using single-cell RNA sequencing (scRNA-seq) data. However, few studies have evaluated the performance of DGE methods-particularly those implemented in the widely used Seurat package. In this study, we systematically assessed 10 DGE methods available in Seurat-Wilcox, Wilcox-limma, bimod, roc, t, negbinom, Poisson, LR, MAST, and DESeq2-using simulated and real scRNA-seq datasets. We evaluated each method's performance across varying sample sizes, levels of sparsity, and proportions of truly differentially expressed genes. Metrics assessed included false discovery rate (FDR), sensitivity, specificity, accuracy, area under the receiver operating characteristic curve (AUC), and area under the precision-recall curve (AUPRC). Among all methods, DESeq2 consistently demonstrated the best overall performance, showing the highest AUC and AUPRC across all tested conditions. Based on our findings, we recommend DESeq2 as the preferred method for DGE analysis in scRNA-seq data.