Table 2. Clustering methods.

Method	Description	Reference
ascend (v0.5.0)	PCA dimension reduction (dim=30) and iterative hierarchical clustering	36
CIDR (v0.1.5)	PCA dimension reduction based on zero-imputed similarities, followed by hierarchical clustering	37
FlowSOM (v1.12.0)	PCA dimension reduction (dim=30) followed by self-organizing maps (5×5, 8×8 or 15×15 grid, depending on the number of cells in the data set) and hierarchical consensus meta-clustering to merge clusters	38
monocle (v2.8.0)	t-SNE dimension reduction (initial PCA dim=50, t-SNE dim=3) followed by density-based clustering	25, 39
PCAHC	PCA dimension reduction (dim=30) and hierarchical clustering with Ward.D2 linkage	33, 40
PCAKmeans	PCA dimension reduction (dim=30) and K-means clustering with 25 random starts	33, 41
pcaReduce (v1.0)	PCA dimension reduction (dim=30) and k-means clustering through an iterative process. Stepwise merging of clusters by joint probabilities and reducing the number of dimensions by PC with lowest variance. Repeated 100 times followed consensus clustering using the clue package	42
RaceID2 (March 3, 2017 version)	K-medoids clustering based on Pearson correlation dissimilarities	43
RtsneKmeans	t-SNE dimension reduction (initial PCA dim=50, t-SNE dim=3, perplexity=30) and K-means clustering with 25 random starts	34, 41, 44
SAFE (v2.1.0)	Ensemble clustering using SC3, CIDR, Seurat and t-SNE + Kmeans	45
SC3 (v1.8.0)	PCA dimension reduction or Laplacian graph. K-means clustering on different dimensions. Hierarchical clustering on consensus matrix obtained by K-means	46
SC3svm (v1.8.0)	Using SC3 to derive the clusters for half of the cells, then using a support vector machine (SVM) to classify the rest	46, 47
Seurat (v2.3.1)	Dimension reduction by PCA (dim=30) followed by nearest neighbor graph clustering	17
TSCAN (v1.18.0)	PCA dimension reduction followed by model-based clustering	48