Abstract
This study examines the noise and biases introduced by technical factors in single-cell RNA sequencing (scRNA-seq) data, presenting a thorough benchmarking analysis of six widely utilized normalization methods. The evaluation of these methods is conducted from three perspectives: cell clustering, differential expression analysis, and computational resource requirements, utilizing seven real datasets alongside four simulated datasets. The findings indicate that Dino excels in clustering 10 × datasets and those with a substantial number of cells, while scTransform demonstrates strong performance with datasets produced through full-length library preparation protocols. Additionally, SCnorm is identified as suitable for small-scale datasets. This research serves as a significant reference for scholars in selecting appropriate normalization tools, thereby enhancing the accuracy and reliability of subsequent analyses of scRNA-seq data.