A posterior probability based Bayesian method for single-cell RNA-seq data imputation

•We propose BayesImpute, a novel statistical algorithm to impute scRNA-seq data. BayesImpute first identifies likely dropouts, and then only imputes these values, which preserves the true biological signal and reduces the introduction of unwanted bias during imputation.•Unlike other statistical imputation methods, the identification process of BayesImpute is straightforward and avoids parameter estimation, which increases its efficiency and user-friendliness in practice. In addition, BayesImpute takes advantage of cell-to-cell relationships and employs the Bayes approach to recover true biological signals, making the obtained estimations interpretable.•Using simulated and real scRNA-seq datasets, we demonstrate that BayesImpute can effectively identify dropouts, reduce the introduction of false positive signals, better recover the missing biological signals, and improve the reliability of some downstream analyses. More importantly, BayesImpute outperforms other statistical model-based methods in terms of computational running time, memory usage, and scalability.