M3C {M3C}R Documentation

M3C: Monte Carlo Reference-based Consensus Clustering

Description

This is the M3C core function, which is a reference-based consensus clustering algorithm. The basic idea is to use a multi-core enabled Monte Carlo simulation to drive the creation of a null distribution of stability scores. The Monte Carlo simulations maintains the feature correlation structure of the input data. Then the null distribution is used to compare the reference scores with the real scores and an empirical p value is calculated for every value of K to test the null hypothesis K=1. We derive the Relative Cluster Stability Index (RCSI) as a metric for selecting K, which is based on a comparison against the reference mean. A faster alternative is included that includes a penalty term to prevent overfitting, called the Penalised Cluster Stability Index (PCSI).

Usage

M3C(mydata, cores = 1, iters = 100, maxK = 10, des = NULL,
  ref_method = c("reverse-pca", "chol"), repsref = 100,
  repsreal = 100, clusteralg = c("pam", "km", "spectral", "hc"),
  distance = "euclidean", pacx1 = 0.1, pacx2 = 0.9,
  printres = FALSE, printheatmaps = FALSE, showheatmaps = FALSE,
  seed = NULL, removeplots = FALSE, dend = FALSE, silent = FALSE,
  doanalysis = FALSE, analysistype = c("survival", "kw", "chi"),
  variable = NULL, fsize = 18, method = 1)

Arguments

mydata

Data frame or matrix: Contains the data, with samples as columns and rows as features

cores

Numerical value: how many cores to split the monte carlo simulation over

iters

Numerical value: how many Monte Carlo iterations to perform (default: 100, recommended: 5-200)

maxK

Numerical value: the maximum number of clusters to test for, K (default: 10)

des

Data frame: contains annotation data for the input data for automatic reordering

ref_method

Character string: refers to which reference method to use (recommended: leaving as default)

repsref

Numerical value: how many resampling reps to use for reference (default: 100, recommended: 100-250)

repsreal

Numerical value: how many resampling reps to use for real data (default: 100, recommended: 100-250)

clusteralg

String: dictates which inner clustering algorithm to use for M3C

distance

String: dictates which distance metric to use for M3C (recommended: leaving as default)

pacx1

Numerical value: The 1st x co-ordinate for calculating the pac score from the CDF (default: 0.1)

pacx2

Numerical value: The 2nd x co-ordinate for calculating the pac score from the CDF (default: 0.9)

printres

Logical flag: whether to print all results into current directory

printheatmaps

Logical flag: whether to print all the heatmaps into current directory

showheatmaps

Logical flag: whether to show the heatmaps on screen

seed

Numerical value: fixes the seed if you want to repeat results, set the seed to 123 for example here

removeplots

Logical flag: whether to remove all plots

dend

Logical flag: whether to compute the dendrogram and p values for the optimal K or not

silent

Logical flag: whether to remove messages or not

doanalysis

Logical flag: whether to analyse the clinical variable supplied (univariate only)

analysistype

Character string: refers to which kind of statistical analysis to do on the data, survival, Kruskal-Wallis (kw), or chi-squared (chi)

variable

Character string: if not doing survival what is the dependant variable (column name) called in the data frame

fsize

Numerical value: determines the font size of the ggplot2 plots

method

Numerical value: 1 refers to the Monte Carlo reference procedure, 2 to using a penalty term (faster)

Value

A list, containing: 1) the stability results and 2) all the output data (another list) 3) reference stability scores (see vignette for more details on how to easily access)

Examples

res <- M3C(mydata)

[Package M3C version 1.6.0 Index]