Get clustering with SNN graph — runScranSNN • singleCellTK

Perform SNN graph clustering on a SingleCellExperiment object, with graph construction by buildSNNGraph and graph clustering by "igraph" package.

runScranSNN(
  inSCE,
  useReducedDim = NULL,
  useAssay = NULL,
  useAltExp = NULL,
  altExpAssay = "counts",
  altExpRedDim = NULL,
  clusterName = "scranSNN_cluster",
  k = 8,
  nComp = 10,
  weightType = c("rank", "number", "jaccard"),
  algorithm = c("louvain", "leiden", "walktrap", "infomap", "fastGreedy", "labelProp",
    "leadingEigen"),
  BPPARAM = BiocParallel::SerialParam(),
  seed = 12345,
  ...
)

Arguments

inSCE: A SingleCellExperiment object.
useReducedDim: A single character, specifying which low-dimension representation (reducedDim) to perform the clustering algorithm on. Default NULL.
useAssay: A single character, specifying which assay to perform the clustering algorithm on. Default NULL.
useAltExp: A single character, specifying the assay which altExp to perform the clustering algorithm on. Default NULL.
altExpAssay: A single character, specifying which assay in the chosen altExp to work on. Only used when useAltExp is set. Default "counts".
altExpRedDim: A single character, specifying which reducedDim within the altExp specified by useAltExp to use. Only used when useAltExp is set. Default NULL.
clusterName: A single character, specifying the name to store the cluster label in colData. Default "scranSNN_cluster".
k: An integer, the number of nearest neighbors used to construct the graph. Smaller value indicates higher resolution and larger number of clusters. Default 8.
nComp: An integer. The number of components to use for graph construction. Default 10. See Detail.
weightType: A single character, that specifies the edge weighing scheme when constructing the Shared Nearest-Neighbor (SNN) graph. Choose from "rank", "number", "jaccard". Default "rank".
algorithm: A single character, that specifies the community detection algorithm to work on the SNN graph. Choose from "leiden", "louvain", "walktrap", "infomap", "fastGreedy", "labelProp", "leadingEigen". Default "louvain". See Detail.
BPPARAM: A BiocParallelParam object to use for processing the SNN graph generation step in parallel.
seed: Random seed for reproducibility of results. Default NULL will use global seed in use by the R environment.
...: Other optional parameters passed to the igraph clustering functions. See Details.

Value

The input SingleCellExperiment object with factor cluster labeling updated in colData(inSCE)[[clusterName]].

Details

Different graph based clustering algorithms have diverse sets of parameters that users can tweak. The help information can be found here:

for "louvain", see function help cluster_louvain
for "leiden", see function help cluster_leiden
for "walktrap", see function help cluster_walktrap
for "infomap", see function help cluster_infomap
for "fastGreedy", see function help cluster_fast_greedy
for "labelProp", see function help cluster_label_prop
for "leadingEigen", see function help cluster_leading_eigen

The Scran SNN building method can work on specified nComp components. When users specify input matrix by useAssay or useAltExp + altExpAssay, the method will generate nComp components and use them all. When specifying useReducedDim or useAltExp + altExpRedDim, this function will subset the top nComp components and pass them to the method.

References

Aaron Lun and et. al., 2016

Examples

data("mouseBrainSubsetSCE")
mouseBrainSubsetSCE <- runScranSNN(mouseBrainSubsetSCE,
                                   useReducedDim = "PCA_logcounts")
#> Tue Jun 28 22:07:24 2022 ... Running 'scran SNN clustering' with 'louvain' algorithm
#> Tue Jun 28 22:07:24 2022 ...   Identified 2 clusters