Table 2. Algorithms used to plot cell conversion trajectories.

Algorithm	Required inputa (optional in parentheses)	Dimension reduction (D)	Trajectory plotting (T)	Cell Clustering (C)	Order	Example application	Properties	Assumptions / requirementsb
Monocle, 2014 [45]	Branches	ICA	Weighted complete graph, MST	No	D,T	scRNA-seq datasets for human myoblast differentiation	Robust to changes in subpopulation structure, subsampling. Resolves cellular transitions during differentiation through temporal profiling of the entire transcriptome without a priori knowledge of marker genes.	Continuous transcriptome path. Known number of branches.
SCUBA, 2014 [72]	(Time course, marker genes)	k-means clustering	Gap statistic, penalized likelihood function, cusp bifurcation theory	k-means clustering	DC (simultaneous), T	RT-PCR and scRNA-seq datasets for early mouse embryo development	Robust to experimental platform differences. Uses temporal information.	Continuous transcriptome path. One or two branches.
Wanderlust, 2014 [73]	Starting cells	KNN graph	Sets of random reference points (waypoints) and determines the position of each cell by weighted shortest-path distance	No	D,T	sc-mass cytometry data for human naïve B cell differentiation	Robust to technical and biological noise.	Continuous transcriptome path. Non-branching trajectory. Known starting point of development.
Waterfall, 2015 [74]	None	PCA	k-means clustering, MST	Unsupervised hierarchical clustering	C,D,T	scRNA-seq datasets for adult neurogenesis in mouse hippocampus	Does not need temporal information or a priori knowledge of marker genes. Applicable for diverse single-cell multi-dimensional datasets, including RNA-seq and mass cytometry.	Continuous transcriptome path.
destiny, 2016 [52]	None	KNN graph	Diffusion maps	No	D,T	scRNA-seq datasets for mouse embryonic fibroblast reprogramming; qRT-PCR data for mouse embryonic cell development; sc-mass cytometry for mouse induced pluripotent stem cell reprogramming	Robust to biological noise and variation in sample density.	Continuous transcriptome path.
DPT, 2016 [53]	(Marker genes)	Diffusion maps	Diffusion maps, branching identification by comparing two independent diffusion pseudo-time (DPT) orderings over cells, metastable state identification	No	DT (simultaneous)	scRNA-seq datasets for mouse blood cell development	Robust to parameter choice. Does not need temporal information, a priori knowledge of marker genes, or starting and end cell identities.	Continuous and smooth transcriptome path.
SLICE, 2016 [48]	(Cell grouping, marker genes)	PCA	Linear Prize-Collecting Steiner Tree (LPCST) problem, MST, shortest-path approach, principal curve based approach	Partitioning around medoids (PAM) / complete weighted graph	D,C,T	scRNA-seq datasets for differentiation of mouse lung alveolar type	Robust to parameter choice. Does not need temporal information, a priori knowledge of marker genes, or starting and end cell identities.	Continuous transcriptome path. Cells with higher pluripotent potential are hypothesized to express genes with more diverse and heterogeneous functions.
SLICER, 2016 [75]	None	KNN graph, locally linear embedding (LLE)	Geodesic entropy	No	D,T	scRNA-seq datasets for mouse lung and neural cells	Robust to biological noise and presence of irrelevant elements (genes). Detects non-tree-like loop structures in development path. Does not need temporal information or a priori knowledge of marker genes.	Continuous transcriptome path.
TSCAN, 2016 [61]	None	PCA / ICA	MST	Hierarchical clustering	D,C,T	scRNA-seq datasets for human skeletal muscle myoblast differentiation	Graphical user interface. Direct comparison with other algorithms.	Continuous transcriptome path. Known number of cell clusters.
Wishbone, 2016 [76]	Starting and ending cells, (marker genes)	Diffusion maps	KNN graph, waypoint sparse approximation	No	D,T	sc-mass cytometry data and scRNA-seq data for human myeloid differentiation	Robust to parameter choice. Good branching point detection in bifurcating systems.	Continuous transcriptome path. One or two branches.
Monocle 2, 2017 [60]	(Number of cell fates)	PCA / t-SNE / diffusion maps	Reversed graph embedding (RGE)	k-means clustering	D,C,T	scRNA-seq data for human myoblast differentiation	Robust to biological noise. Does not need a priori knowledge of genes that characterize the biological process or the number of branch points in the trajectory.	Continuous transcriptome path.
scTDA, 2017 [54]	(Time course)	MDS, top 5000 variant genes	Single-cell topological data analysis (scTDA)	Single-linkage clustering	D,C,T	scRNA-seq data for mouse motor neuron differentiation	Detects transient cellular populations and their transcriptional repertoires. Detects non-tree-like loop structures in development path. Identifies cell-cycle-related features from loop structures in the trajectory.	Continuous transcriptome path.
CellRouter, 2018 [47]	(Marker genes)	t-SNE / diffusion maps	Flow network	KNN graph	D,C,T	scRNA-seq data for human neutrophil differentiation	Robust to subpopulation structure, subsampling, and choice of dimension reduction techniques.	A continuum of phenotypically distinct subpopulations. State transitions are continuous with molecular hallmarks activated or silenced in a progressive manner.
Slingshot, 2018 [50]	(Starting and ending cells)	PCA / ICA / diffusion maps	MST	k-means clustering / Gaussian mixture modeling	D,C,T	scRNA-seq data for olfactory stem cell niche	Robust to subsampling and cluster assignments. Flexibility in upstream analysis, including choice of dimension reduction and clustering algorithms. Identification of multiple cell fates.	Continuous transcriptome path.

Abbreviations: independent component analysis (ICA), minimum spanning tree (MST), k-nearest neighbors (KNN), principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE), multidimensional scaling (MDS)

^aThese are in addition to required gene and/or protein expression data.

^bAlthough not always explicitly mentioned as a criterion, a continuous transcriptome path is an implicit assumption for all algorithms.