clustering_analysis: Clustering, internal evaluation and batch effect estimation

In vittoriofortino84/COPS: Clustering algorithms for Omics-driven Patient Stratification

Clustering, internal evaluation and batch effect estimation

Description

Single embedding or dataset evaluation

Usage

clustering_analysis(
  dat,
  n_clusters = 2:5,
  cluster_methods = c("hierarchical", "diana", "kmeans"),
  clustering_dissimilarity = NULL,
  distance_metric = "euclidean",
  correlation_method = "spearman",
  hierarchical_linkage = "complete",
  kmeans_num_init = 100,
  kmeans_max_iters = 100,
  kmeans_tol = 1e-08,
  gmm_modelNames = NULL,
  gmm_shrinkage = 0.01,
  knn_neighbours = 30,
  knn_jaccard = TRUE,
  kernel = "linear",
  kernel_gamma = 1,
  kernel_center = TRUE,
  kernel_normalize = TRUE,
  kkmeans_algorithm = "spectral",
  kkmeans_refine = FALSE,
  kkmeans_maxiter = 100,
  kkmeans_n_init = 100,
  kkmeans_tol = 1e-08,
  ...
)

Arguments

dat	A data.frame with features on columns labeled as "dim[0-9]+", must also contain "id" column.
n_clusters	A vector of integers defining the number of clusters.
cluster_methods	A vector of clustering method names, see details for options.
clustering_dissimilarity	A dissimilarity matrix used in some methods such as hierarchical clustering. Computed with clustering_dissimilarity_from_data if missing.
distance_metric	Either "euclidean" or "correlation".
correlation_method	Method for cor.
hierarchical_linkage	See flashClust.
kmeans_num_init	See KMeans_rcpp.
kmeans_max_iters	See KMeans_rcpp.
kmeans_tol	See KMeans_rcpp.
gmm_modelNames	Sepcifies model type for Mclust
gmm_shrinkage	Shrinkage parameter for priorControl.
knn_neighbours	number of nearest neighbours for community detection.
knn_jaccard	computes shared neighbour weights with Jaccard ubdex if TRUE.
kernel	kernel for kernel k-means, options: "linear", "gaussian", "rbf", "jaccard", "tanimoto"
kernel_gamma	gamma for the Gaussian/RBF kernel, higher values correspond to more complicated boundaries
kernel_center	center kernels if TRUE
kernel_normalize	normalize kernels to L2 norm 1 if TRUE
kkmeans_algorithm	See kernel_kmeans options.
kkmeans_refine	See kernel_kmeans.
kkmeans_maxiter	maximum number of iterations for kernel k-means
kkmeans_n_init	number of random initializations for kernel k-means++
kkmeans_tol	delta error convergence threshold for spectral clustering
...	extra arguments are ignored

Details

Supported clustering methods are:

• "hierarchical" - agglomerative hierarchical clustering

• "diana" - divisive hierarchical clustering analysis

• "kmeans" - k-means++

• "model" - Gaussian Mixture Models

• "knn_communities" - Louvain community detection on shared k nearest neighbour graphs

• "spectral" - spectral clustering

• "SC3" - consensus clustering http://dx.doi.org/10.1038/nmeth.4236, note that this requires SC3 installation which is not required by default

• "kkmeans" - kernelized k-means initialized by a spectral approximation

• "kkmeanspp" - kernelized k-means++ with random initializations

Value

Returns a list containing clusters, metrics, and chisq.test p-values if batch_label was supplied

vittoriofortino84/COPS documentation built on Jan. 28, 2025, 3:16 p.m.