cluster_sim_spectrum: Calculate Cluster Similarity Spectrum

In quadbiolab/simspec: Calculate data representation of Reference/Cluster Similarity Spectrum

Calculate Cluster Similarity Spectrum

Description

Calculate Cluster Similarity Spectrum (CSS), given expression of the data and cell labels used to distinguish samples. Clustering is applied to cells of each sample separately, similarities of one cell to those clusters are calculated and normalized.

Usage

## Default S3 method:
cluster_sim_spectrum(
  object,
  labels,
  cluster_labels = NULL,
  dr = NULL,
  dr_input = NULL,
  num_pcs_compute = 50,
  num_pcs_use = 20,
  redo_pca = FALSE,
  k = 20,
  min_batch_size = k * 2,
  ...,
  cluster_method = c("Seurat", "walktrap"),
  cluster_resolution = 0.6,
  min_cluster_num = 3,
  spectrum_type = c("corr_ztransform", "corr_kernel", "corr_raw", "nnet", "lasso"),
  corr_method = c("spearman", "pearson"),
  use_fast_rank = TRUE,
  lambda = 50,
  threads = 1,
  train_on = c("raw", "pseudo", "rand"),
  downsample_ratio = 1/10,
  k_pseudo = 10,
  logscale_likelihood = F,
  merge_spectrums = FALSE,
  merge_height_prop = 1/10,
  spectrum_dist_type = c("pearson", "euclidean"),
  spectrum_cl_method = "complete",
  return_css_only = T,
  verbose = T
)

## S3 method for class 'Seurat'
cluster_sim_spectrum(
  object,
  label_tag,
  cluster_col = NULL,
  var_genes = NULL,
  use_scale = F,
  use_dr = "pca",
  dims_use = 1:20,
  redo_pca = FALSE,
  redo_pca_with_data = FALSE,
  k = 20,
  min_batch_size = k * 2,
  ...,
  cluster_resolution = 0.6,
  spectrum_type = c("corr_ztransform", "corr_kernel", "corr_raw", "nnet", "lasso"),
  corr_method = c("spearman", "pearson"),
  lambda = 50,
  threads = 1,
  train_on = c("raw", "pseudo", "rand"),
  downsample_ratio = 1/10,
  k_pseudo = 10,
  logscale_likelihood = F,
  merge_spectrums = FALSE,
  merge_height_prop = 1/10,
  spectrum_dist_type = c("pearson", "euclidean"),
  spectrum_cl_method = "complete",
  reduction.name = "css",
  reduction.key = "CSS_",
  return_seuratObj = T,
  verbose = T
)

cluster_sim_spectrum(object, ...)

Arguments

object	An object
labels	Labels specifying different samples
cluster_labels	Use the provided clustering results instead of doing clustering per sample
dr	Dimension reduction matrix used for clustering. When it is NULL, truncated PCA is run on the expression matrix for dimension reduction
dr_input	Alternative expression matrix used for dimension reduction. Ignore if dr is specified
num_pcs_compute	Number of PCs to calculate. Ignore if dr is specified
num_pcs_use	Number of PCs used for clustering
redo_pca	If TRUE, PCA is rerun for each sample separately for clustering
k	Number of nearest neighbors of the kNN network used for clustering
min_batch_size	The minimal cell number of a batch to be clustered to generate references
...	Other parameters to build_knn_graph
cluster_method	Method used to apply clustering to the kNN network. By default it calls FindClusters in Seurat using Louvain method. Alternative method is the walktrap community identification algorithm in igraph
cluster_resolution	Resolution of clustering. Ignore if cluster_method is not Seurat
min_cluster_num	The minimal number of clusters to include a sample in the ref profile (default=3)
spectrum_type	Method to normalize similarities. "corr_ztransform" uses z-transform; "corr_kernel" introduces correlation kernel to convert similarities to likelihood; "corr_raw" uses no normalization; "nnet" and "lasso" build probabilistic prediction model on the data and estimate likelihoods
corr_method	Type of correlation. Ignore if spectrum_type is "nnet" or "lasso"
use_fast_rank	When the presto package is available, use its rank_matrix function to rank sparse matrix
lambda	Lambda in the correlation kernel
threads	Number of threads to use. Only useful if spectrum_type is "lasso"
train_on	Type of data used to train the likelihood model. Only useful if spectrum_type is "nnet" or "lasso"
downsample_ratio	Downsample rate. Only useful if train_on is "pseudo" or "rand"
k_pseudo	Number of nearest neighbors used to construct pseudocells. Only useful if train_on is "pseudo"
logscale_likelihood	If TRUE, estimated likelihoods are log-transformed. Ignore if spectrum_type is "corr_ztransform" or "corr_raw"
merge_height_prop	The height of dendrogram to cut. Ignore if merge_spectrum is FALSE
spectrum_dist_type	Type of distance to construct the dendrogram of spectrums. Ignore if merge_spectrum is FALSE
spectrum_cl_method	Method of hierarchical clustering to construct the dendrogram of spectrums. Ignore if merge_spectrum is FALSE
return_css_only	If FALSE, not only the calculated CSS matrix, but also other information to recalculate the spectrum is returned
verbose	If TRUE, progress message is provided
label_tag	Column in the meta.data slot showing sample labels
cluster_col	Column in the meta.data slot showing the cluster labels
var_genes	Genes used for similarity calculation. If NULL, predefined variable features are used
use_scale	If TRUE, scale.data rather than data slot is used for similarity calculation
use_dr	Name of reduction used for clustering
dims_use	Dimensions in the reduction used for clustering
redo_pca_with_data	If TRUE, data slot is used to redo PCA for each sample. Ignore if redo_pca is FALSE
reduction.name	Reduction name of the CSS representation in the returned Seurat object
reduction.key	Reduction key of the CSS representation in the returned Seurat object
return_seuratObj	If TRUE, a Seurat object with CSS added as one dimension reduction representation is returned. Otherwise, a list with CSS matrix and the calculation model is returned
merge_spectrum	If TRUE, similar similarity spectrums are averaged

quadbiolab/simspec documentation built on March 8, 2024, 11:59 p.m.