Inferring phenotypes of single cells based on the expression profiles of phenotype-associated marker genes in bulks and single cells

Single-cell transcriptomes is advantageous in characterizing cell types and states, but not sufficient in describing cell phenotypes due to small cohorts in single-cell datasets. Thus, inferring single-cell phenotypes with the leverage of bulk data is practical. Here we proposed an algorithm for Single Cells Phenotype Prediction based on the expression profiles of phenotype-associated marker genes in bulks and single cells (ScPP). ScPP first analyzes bulk data to identify phenotype-associated marker genes. Next, ScPP evaluates the enrichment scores of the phenotype-associated marker gene sets in single cells using the AUCell algorithm. The single cells with the phenotype are defined as the intersection of the single cells with top ranks according to the phenotype-associated AUC values and the single cells with bottom ranks according to the opposite phenotype-associated AUC values. Finally, all single cells are determined as phenotype+, phenotype- or background. We demonstrated that ScPP could effectively recognize cell subpopulations with specific phenotypes, including tumor malignancy, estrogen receptor status, microsatellite instability, copy number variation, survival and immunotherapy response. Compared to the established algorithms (Scissor and scAB), ScPP displays more excellent predictive performance and needs less running time. Thus, ScPP is an effective and competitive tool for inferring cell phenotypes.