Fig. 4
From: A geometric approach to characterize the functional identity of single cells
ACTION Kernel Robustness. A series of expression profiles with varying degrees of dropout has been simulated from the CellLines dataset. In each case, we compute different metrics and use kernel k-means to identify cell types. The quality of cell-type identification is assessed with respect to known annotation from the original paper using three different extrinsic measures: a Adjusted Rand Index (ARI), b F-score, and c Normalized Mutual Information (NMI). These results show that ACTION and MDS have the most stable performance over dropout. Error bars correspond to repeated samples of perturbed expression profiles
