Fig. 1
From: CIDR: Ultrafast and accurate clustering through imputation for single-cell RNA-seq data
A toy example to illustrate the effect of dropouts in scRNA-seq data on clustering and how CIDR can alleviate the effect of dropouts. a This toy example consists of eight single cells divided into two clusters (the red cluster and the blue cluster). Dropout causes the within-cluster distances among the single cells in the red cluster to increase dramatically, as well as increasing the between-cluster distances between single cells in the two clusters. bCIDR reduces the dropout-induced within-cluster distances while largely maintaining the BC distances. c The hierarchical clustering results using the original data set (no dropout), the dropout-affected data set, and the dropout-affected data set analyzed using CIDR. BC between clusters, DO dropout, scRNA-seq single-cell RNA-seq, WC within clusters. d Using a step function W(x) to estimate the real dropout rate function P(x), we can show that CIDR always shrinks the expected distance between any two points (x1 and x2), and that the expected shrinkage rate is higher for those pairs of points that are closer together