README

Clustering by directions in CA space

This package implements a clustering algorithms that determines clusters iteratively by their directions in CA space. Unlike most other algorithms it does not require prior knowledge of the number of clusters in the data, but can instead infer them during clustering. The package can be installed through GitHub:

devtools::install_github("ClemensKohl/CAdir")

Note that currently you have to also install CAbiNet from GitHub for the package to function.

Quick start

Download example data and perform basic preprocessing:

suppressPackageStartupMessages({
  library(CAdir)
  library(APL)

  # packages for loading the example:
  library(scRNAseq)
  library(scran)
  library(scater)
  library(scuttle)
})

Warning: replacing previous import 'S4Arrays::makeNindexFromArrayViewport' by
'DelayedArray::makeNindexFromArrayViewport' when loading 'SummarizedExperiment'

Warning: replacing previous import 'S4Arrays::makeNindexFromArrayViewport' by
'DelayedArray::makeNindexFromArrayViewport' when loading 'HDF5Array'

set.seed(2358)

sce <- scRNAseq::ZeiselBrainData()
clust <- scran::quickCluster(sce)
sce <- scran::computeSumFactors(sce, cluster = clust, min.mean = 0.1)
sce <- scuttle::logNormCounts(sce)
dec <- scran::modelGeneVar(sce)
top_genes <- scran::getTopHVGs(dec, prop = 0.8)
sce <- sce[top_genes, ]

Correspondence analysis

cnts <- as.matrix(logcounts(sce))

ca <- cacomp(obj = cnts,
             princ_coords = 3,
             dims = 30,
             top = nrow(cnts),
             residuals = "pearson",
             python = TRUE)

cell_types <- sce$level1class
cat("Number of cell types:", length(unique(cell_types)), "\n")

Number of cell types: 7 

CAdir

cadir <- dirclust_splitmerge(
  caobj = ca,
  k = 10,
  cutoff = NULL,
  min_cells = 20,
)

Inferred cutoff angle: 66.85

Iteration 1
    Merging cluster_1 with cluster_4
•   Merging cluster_1 with cluster_9
Iteration 2
Iteration 3
Iteration 4
Iteration 5

cadir

caclust object with 3005 cells and 1518 genes.
8 clusters found.
Clustering results:

 cluster   ncells ngenes
 cluster_1 1225     2   
 cluster_2   92   233   
 cluster_3  310    43   
 cluster_4   37   215   
 cluster_5  721    22   
 cluster_6  242   148   
 cluster_7  157   491   
 cluster_8  221   364   

Annotate cell clusters:

cadir <- annotate_biclustering(
  obj = cadir,
  universe = rownames(sce),
  org = "mm"
)
cadir

caclust object with 3005 cells and 1518 genes.
8 clusters found.
Clustering results:

 cluster          ncells ngenes
 Wnt2+_cell       1225     2   
 Mural_cell         92   233   
 CCK_basket_cell   310    43   
 Ciliated_cell      37   215   
 cluster_5         721    22   
 Astrocyte         242   148   
 Macrophage        157   491   
 Endothelial_cell  221   364   

Rank cluster specific genes:

cadir <- rank_genes(cadir = cadir, caobj = ca)
top <- top_genes(cadir)

# Top genes for cluster Macrophage
head(top[top$Cluster == "Macrophage", ])

                  Rowname    Score Row_num    Cluster
Macrophage.Fcgr3    Fcgr3 3.345201      14 Macrophage
Macrophage.Fcer1g  Fcer1g 3.338749      31 Macrophage
Macrophage.Fcrls    Fcrls 3.301021      15 Macrophage
Macrophage.Emr1      Emr1 3.246037      52 Macrophage
Macrophage.Tyrobp  Tyrobp 3.129216      10 Macrophage
Macrophage.C1qc      C1qc 3.126874      24 Macrophage

Plot results

cluster_apl(cadir = cadir,
            caobj = ca,
            direction = cadir@directions["Macrophage",],
            group = which(cadir@cell_clusters == "Macrophage"),
            cluster = "Macrophage",
            show_genes = TRUE,
            label_genes = TRUE)

plot_clusters(
  cadir = cadir,
  caobj = ca,
  show_genes = TRUE,
  title_prefix = "",
  axis = TRUE
)

sm_plot(
  cadir = cadir,
  caobj = ca,
  show_genes = FALSE,
  show_cells = TRUE,
  annotate_clusters = TRUE,
  highlight_cluster = TRUE,
  org = "mm"
)

Package settings

Verbosity of the messages can be controlled with rlang. To turn all messages off:

rlang::local_options(mypackage.verbose = "quiet")

To turn them back on:

rlang::local_options(mypackage.verbose = "verbose")