Finding Significantly Enriched Cells in Single-Cell RNA Sequencing by Single-Sample Approaches

Gene set analysis is a leading bioinformatical technique allowing comparison of phenotypes on gene set level, which is applied to different transcriptome-wide gene expression platforms and omics levels. The aim of this study was to measure the performance of three single-sample gene set enrichment algorithms, based on their ability to obtain the statistical significance of enrichment in each cell separately using scRNA-Seq data. The peripheral blood mononuclear cell dataset was used in the evaluation process and individual enrichment within the B cell subtype was investigated based on reference gene set collection. Sensitivity, specificity, prioritization, and balanced accuracy were used as evaluation metrics, accompanied by correlation analysis between gene sets. AUCell, originally designed for scRNA-Seq, showed the best sensitivity and balanced accuracy, good prioritization and acceptable specificity. However, large correlation between gene set size and specificity was observed, so we recommend its usage on large gene sets (>80). Moreover, the computational time is much longer compared to other tested methods. Among other algorithms, CERNO gave very high specificity and prioritization, but the sensitivity needs to be enhanced by algorithm improvement. Finally, the problem of the “gold standard” dataset and gene set collection that could be used for gene set analysis algorithms performance evaluation in scRNA-Seq, was stated and the initial solution was presented.