Advances in single cell RNA-seq (scRNA-seq) technology has enabled identification and quantification of cell type composition of complex tissues across multiple samples. It is increasingly common to ask questions such as ’Is the proportion of this cell type significantly altered between conditions?’ In other words, we want to perform differential proportion analysis on the cell count matrix based on scRNA-seq data to identify cell types that have a significant increase or decrease in proportion. Our initial work showed that a simple statistical test such as Fisher’s exact test produces high false positive rate, suggesting that there are additional variability beyond random sampling. We reason that a possible source of variation is cell-type mis-classification, which can be estimated by cell-to-cell similarity matrix computed during the clustering process. We implemented our idea in an R package and tested it using simulation and real scRNA-seq data sets. In this seminar we will introduce our method and share our initial evaluation results.