Abhishek Narain Singh*, Krishan Pal
There is a marked technical variability and a high amount of missing observations in the single-cell data that we obtain from experiments. Apart from that clearly each of the batch of experiments have a batch effect on every cell in the batch. This batch effect can be taken into advantage for dealing with imputation, given that all the cells in a given batch belong to the same tissue. Here we introduce ‘BaySiCle’, a novel Bayesian inference based method combined with k-nearest neighbor’s algorithm for the imputation of missing data in scRNA-seq counts. The priors are found out based on expression value across cells for all the single cells of the same batch. We demonstrate using sample scRNA-seq datasets and simulated expression data that BaySiCle allows robust imputation of missing values generating realistic transcript distributions that match single molecule fluorescence in situ hybridization measurements. By using priors as obtained by the dataset structures in the not just the experimental set-up batch, but also the same group of cells, BaySiCle improves accuracy of imputation to be that much closer to its similar alternatives.