Overview BayesImpute is a statistical model that first identifies likely dropout events through the expression rate and coefficient of variation of gene in cell subpopulation, then imputes these values by constructing posterior distribution for each gene, and treats posterior mean as recovered expression value. As shown in Figure, the workflow of BayesImpute algorithm consists of four steps. The first step is preprocessing, which includes gene filtering, expression normalization and logarithmization. The second step is dimensionality reduction and preliminary clustering, which selects similar cells for subsequent identification and imputation step. During the identification step, BayesImpute distinguishes between dropout events and true zeros. Doing so reduces unwanted biases to the rest data and improves imputation accuracy. Finally, BayesImpute assumes that the actual gene expression values follow Gaussian distribution. It then follows Bayes rule to output the posterior distribution of true expression, and the posterior mean is used as recovered dropout values.