identify_regions: Identify regions of interest

In theislab/kbranches: K-Branches clustering

Description

Identifies regions of interest (the tips of branches, or branching regions) based on comparing the GAP scores#'acquired by running local clustering by 'kbranches.local'. Performs local filtering to reduce noise in the extracted labels.

Usage

identify_regions(input_dat, mode = "tip", tip_mode = "3",
  gap_scores = NULL, smoothing_region = NULL,
  smoothing_region_thresh = NULL, Dist, dotsize1 = 7, dotsize2 = 7,
  nclust = NULL, nclust.max = 5, B.max = 100, SE.factor = 2.58,
  repeats = 20, gapm_tips = F)

Arguments

input_dat:	data frame of input data with rows=samles and cols=dimensions.
mode:	=c('tip','branch') find either the tips or the branching regions
tip_mode:	=c('2','3','both') find the tips using the GAP statistic of 1 vs 2,3 or both
gap_scores:	list, output of function 'kbranches.local' for the same 'input_dat'
smoothing_region:	number of neighbours in to consider for label filtering
smoothing_region_thresh:	minimum number of thresholds with same label in the neighbourhood required for the sample to keep it's label
Dist:	matrix of sample to sample distances
dotsize1:	size of points of class1 (red), used when plotting in 3D
dotsize2:	size of points of class2 (green), used when plotting in 3D
nclust:	number of clusters (tips/branching regions). If left NULL, will be estimated.
nclust.max:	maximum possible number of clusters to consider. Only used if nclust==NULL
B.max:	maximum number of bootstrap datasets used to calculate the GAP statistic to estimate nclust. Only used if nclust==NULL
SE.factor:	used to estimate nclust, argument to cluster::maxSE. Only used if nclust==NULL
repeats:	number of times to estimate nclust (for stability - the nclust most frequently considered 'best' is finally extracted). Only used if nclust==NULL

Value

a list with elements:

• - is_in_region: a logical vector indicating which samples are part of the region (TRUE) and which not (FALSE)

• - is_in_region_filtered: same as 'is_in_region', but after S_neib filtering to reduce noise.

• - cluster: cluster assignment for each of the data points in is_in_region_filtered

• - nclust: number of clusters identified, or specified by the user

• - cluster_frequency: how many times (out of 'repeats') each number of clusters (of the nclust.max) was considered 'best'. NULL if nclust was provided by the user

Examples

#see example of kbranches.local

theislab/kbranches documentation built on Feb. 27, 2020, 11:01 a.m.