Discriminative feature of cells characterizes cell populations of interest by a small subset of genes

Differentially expressed gene-based analysis have faced difficulties associated with the inflation of P-values because of both the large sample size and selection bias introduced by exploratory data analysis in single-cell transcriptomics. Here, we propose the concept of discriminative feature of cells (DFC), implemented by logistic regression with an adaptive LASSO penalty. Adaptive LASSO selects a small subset of genes to discriminate a population of interest. DFC well captured non-independent structure of expression patterns such as cell-type-specific markers, specific composition of gene expression, and subcategories of this cell population. DFC may complement existing methods for interpreting large data sets.