Table 1 Methods compared in the study
From: A comparison framework and guideline of clustering methods for mass cytometry data
Methods | Implementation tools | Description | Ref | |
|---|---|---|---|---|
Unsupervised | Accense | MATLAB | tSNE dimension reduction and 2D projection, kernel-based estimation of density, density-based peak-finding and partitioning | [15] |
PhenoGraph | R (cytofkit package) | Detection of k-nearest neighbors of each cell, Jaccard similarity coefficient as connectivity, community detection based on connection density | [18] | |
Xshift | Vortex | Weighted k-nearest neighbor density estimation, detection of density centroids, cells linked to centroid via density-ascending paths | [21] | |
FlowSOM | R | Self-organizing map (SOM) trained on scaled data, nodes of SOM connected by minimal spanning tree, consensus hierarchical meta-clustering of nodes | [27] | |
flowMeans | R | K estimated by peak numbers of kernel density, kmeans clustering of estimated K, merging clusters by distance metrics | [20] | |
DEPECHE | R | Tuning penalty by resampling dataset, penalized kmeans clustering | [19] | |
kmeans | MATLAB | Standard kmeans procedure | ||
Semi-supervised | ACDC | Python | Marker × cell matrix and cell type × marker table, detect landmark points by community detection, link cells to landmarks by random walkers | [13] |
LDA | MATLAB | Linear discriminant analysis with training datasets | [11] | |