When scientists use single-cell RNA sequencing to study thousands of individual cells, they must first correct the raw data so that meaningful biological differences aren’t obscured by technical quirks. This step, called normalization, adjusts for things like differences in sequencing depth and amplification bias so that comparisons between cells and genes are fair and accurate. However, many different normalization methods exist, and it can be hard to know which one works best for a given dataset.

Researchers at Southeast University in China took on this challenge by comparing six widely used normalization approaches across several real and simulated single-cell datasets. Their goal was to see how each method performed in three key areas: cell clustering, where cells are grouped based on similarity in gene expression; detecting differences in gene expression between groups of cells; and computational efficiency, meaning how much time and memory each method requires.

The results provide practical guidance for researchers working with scRNA-seq data. For datasets generated by high-throughput platforms like 10x Genomics that contain many cells, the method Dino stood out for producing better clustering outcomes. In contrast, scTransform performed especially well on data generated with full-length transcript sequencing protocols. For smaller datasets, SCnorm offered reliable performance.

The benchmark test outcomes of the single-cell RNA sequencing (scRNA-seq)
standardization methods across various simulated datasets.

(a) displays the results of cluster evaluation for each standardization method, where the size of the bubbles corresponds to the value of the evaluation index associated with the vertical coordinate for each dataset, and distinct colors denote different evaluation indices. (b) presents the benchmark test results for differential expression analysis conducted by each standardization method, specifically calculating the F1 scores for both up-regulated and down-regulated differentially expressed genes in comparison to the true differentially expressed genes.

Choosing the right normalization method can make a big difference in downstream analyses such as identifying cell types or detecting genes that are differentially expressed between conditions. By laying out how each method performs under different conditions, this benchmarking work serves as a valuable resource for scientists navigating the complexities of single-cell RNA sequencing analysis.