The paper presents a Bayesian approach to analyzing single-cell RNA sequencing (scRNA-seq) data, which is characterized by high technical noise and intrinsic biological variability. The authors develop a probabilistic model that accounts for the distortions in expression magnitude typical of scRNA-seq measurements, such as "drop-out" events and over-dispersion. This model is used to detect differential expression signatures and identify subpopulations of cells more robustly. The method, called Single Cell Differential Expression (SCDE), incorporates individual cell measurements to estimate the likelihood of gene expression at various levels and the fold change between cell groups. SCDE is evaluated using datasets from mouse embryonic fibroblasts (MEF) and embryonic stem cells (ES), showing higher sensitivity and specificity compared to traditional RNA-seq methods. The paper also introduces modified correlation measures that account for drop-out events, improving the classification of cell populations. Overall, the Bayesian approach provides a more accurate and reliable method for analyzing single-cell data, particularly in complex tissues and diseases.The paper presents a Bayesian approach to analyzing single-cell RNA sequencing (scRNA-seq) data, which is characterized by high technical noise and intrinsic biological variability. The authors develop a probabilistic model that accounts for the distortions in expression magnitude typical of scRNA-seq measurements, such as "drop-out" events and over-dispersion. This model is used to detect differential expression signatures and identify subpopulations of cells more robustly. The method, called Single Cell Differential Expression (SCDE), incorporates individual cell measurements to estimate the likelihood of gene expression at various levels and the fold change between cell groups. SCDE is evaluated using datasets from mouse embryonic fibroblasts (MEF) and embryonic stem cells (ES), showing higher sensitivity and specificity compared to traditional RNA-seq methods. The paper also introduces modified correlation measures that account for drop-out events, improving the classification of cell populations. Overall, the Bayesian approach provides a more accurate and reliable method for analyzing single-cell data, particularly in complex tissues and diseases.