Cross-Organization Single-cell RNA-seq Data Clustering Methods

Single-cell transcriptome sequencing technology provides a new way to understand cell heterogeneity and functional diversity. Batch effect correction and clustering are important steps in cross-tissue single-cell transcriptome data analysis. Choosing the right method has a great impact on downstream analysis such as cross-tissue cell type identification. First, this paper describes five typical batch effect correction methods and nine single-cell clustering methods, and analyzes their advantages and disadvantages. Then, a comparative experimental analysis is conducted on the mouse single-cell transcriptome atlas dataset Tabula Mûris. The performance comparison experiments and analysis of cross-tissue single-cell transcriptome data batch effect correction methods and clustering methods are performed using evaluation indicators such as accuracy, adjusted Rand index, normalized mutual information, and Davidson-Bourdin index. The experimental results show that among the batch effect correction methods, the clustering performance after ComBat batch effect correction is the best, and the accuracy, adjusted Rand index, normalized mutual information, and Davidson-Bourdin index after clustering have obvious advantages over other batch effect correction methods. Among the clustering methods, scDCC and scCCL have the best overall performance in terms of accuracy, adjusted Rand index, normalized mutual information, and Davidson-Bourdin index, with balanced performance and short time consumption.