find_central_elements_by_cluster: Encapsulation of steps to create clusters and determine most...

In Benjamin-Vincent-Lab/binfotron: Binfotron Bioinformatics Analysis Tools Suite

Encapsulation of steps to create clusters and determine most central elements of each cluster

Description

Generate clusters using kmeans method, and determine most representative element for each cluster using a pca analysis (most central feature in pca space) , mhorn similarity index (most similar feature), or pearson/spearman correlation (most correlated feature).

Usage

find_central_elements_by_cluster(
  feature_df,
  anno_mark_font_size = 8,
  annotate_central_elements = T,
  annotate_central_elements_n_clusters = 40,
  central_element_circle_radius = 1/10,
  centrality_methods = "by-rank",
  cluster_id_width = NA,
  cluster_plot_sizes = NA,
  dist_method = "euclidean",
  file_prefix = "central_elements",
  grid_size = 100,
  grid_units = "mm",
  hclust_method = "complete",
  max_clusters = 40,
  max_depth = NA,
  min_clusters = 1L,
  my_threads = 1,
  my_seed = NA,
  output_central_elements = T,
  output_cumulative_variance = F,
  output_dir = ".",
  output_gmt = T,
  output_heatmap = F,
  output_pc1_vs_pc2 = F,
  output_ranked_central_elements = T,
  rank_clm = "Rank",
  rank_df = NULL,
  row_annotation_lwd = 0.25,
  row_annotation_width = 15,
  row_annotation_width_units = "mm",
  row_dend_lwd = 0.25,
  row_dend_width = 15,
  row_dend_width_units = "mm",
  hm_raster_quality = 5,
  show_hm_row_names = T
)

Arguments

feature_df	data.frame on which to perform PCA, mhorn or spearman analysis and kmeans clustering. Importantly: Rows must be named after features.
centrality_methods	A character vector with strings specifying the method for selecting the most central feature of a cluster: two-in-a-row - using PCA, selects the feature that shows up two times in a row as we calculate sum of squares adding more and more PC's is selected max-depth - using PCA, selects the feature with the maximum sum of squares calculated across the number of pc's requested as the "max_depth" first-most-frequent - using PCA, determines the max sum of squares for 2 pcs, 3 pcs, 4 pcs ... up to N pc's and then picks the feature that showed up the most times across all those calculations mhorn - feature most similar to others (ie, largest sum to all other elements) wins spearman - feature most similar to others (ie, largest sum to all other elements) wins pearson - feature most similar to others (ie, largest sum to all other elements) wins by-rank - defaults to the most significant according to rank_df
cluster_id_width	An integer indicating how many characters to use for cluster group and cluster number id's. Defaults to one more than the number of characters in max_clusters.
cluster_plot_sizes	Integer vector indicating which cluster groups to save as plots with clusters circled and central elements labeled. Only used if centrality_methods is one of the pca options.
dist_method	String indicating the method to pass to stats::dist method for clustering
file_prefix	The text to be prepended to the file names for tables and plots
grid_size	Number to specify the size of the heatmap
grid_units	Number to specify the units corresponding to grid_size of the heatmap
hclust_method	String indicating the method to pass to stats::hclust method for clustering
max_clusters	Integer indicating the maximum number of clusters to split data into
max_depth	Integer indicating the maximum depth across principle components to use for determining most central element
min_clusters	Integer indicating the minimum number of clusters to split data into
my_threads	Integer value specifying to number of parallel processes to use when calculating mhorn indices. Defaults to 1.
my_seed	The seed key to use so clustering can be reproduced
output_central_elements	Boolean whether or not to save the table of central elements by cluster group
output_cumulative_variance	Boolean whether to save a plot of the cumulative variance explained by the pca axes. Only used if centrality_methods is one of the pca options.
output_dir	The base directory to which files and plots will be saved
output_gmt	Boolean whether or not to save the gmt data to file
output_heatmap	Boolean whether to save correlation heatmap to file. Ignored if centrality_methods is one of the PCA options.
output_ranked_central_elements	Boolean whether to save the table of unique central elements sorted by rank within cluster group
rank_clm	One-length character vector with the name of the column holding the initial rankings, if any, in either rank_df if one was sent, or in feature_df otherwise
rank_df	Data.frame with feature_df features by row in column one and rank_clm with numeric default ranking for tie-breaking. If <NA> rank_clm will be looked for in feature_df.
output_pc1_vs_pv2	Boolean whether to save a plot of the principle component 1 and 2 axes. Only used if centrality_methods is one of the pca options.

Value

Returns 3 variable list with cluster_members, seed, and results. Results is a named list of each centrality_methods with central_elements and either pca or correlations ( depending on the centrality_methods )

Benjamin-Vincent-Lab/binfotron documentation built on Oct. 1, 2024, 8:33 p.m.