Fig. 2

Validation of JOINT’s clustering performance. a Cell-clustering by JOINT on a simulated dataset with two cell-types and two genes. Scatter plot shows posterior probability (z-axis) for each cell (red dots) belonging to cell-type 1. Expression levels of gene 1 (Dimension 1, Dim 1) and gene 2 (Dimension 2, Dim 2) are shown on the x- and y-axis. b Surface plot shows the probability for individual cells belonging to cell-type 1 (hot color) and 2 (cold color). c - h Comparison of the clustering performance of different algorithms. c Original dataset without dropout (True Labels). d Observed dataset with 0.2 dropout probability. e Cell-clustering by JOINT on the dataset with 0.2 dropout probability. f Cell-clustering by K-means on non-log data with 0.2 dropout probability. g Cell-clustering by K-means on log-transformed data with 0.2 dropout probability. h Cell-clustering by K-mean on Saver-imputed data (non-log) with 0.2 dropout probability. Individual cells in clusters 1 and 2 are shown in red and blue, respectively. i - k The JOINT algorithm determines cell-cluster numbers automatically by likelihood (i), AIC (j), and BIC (k) tests