ISKmeans: IS-Kmeans

In Caleb-Huo/IS-Kmeans: Integrative Sparse Kmeans

Description

Integrative Sparse KMeans

Usage

ISKmeans(d, K = NULL, gamma = NULL, alpha = 0.5, group = NULL,
  nstart = 20, wsPre = NULL, penaltyInfo = NULL, sparseStart = TRUE,
  silent = FALSE, maxiter = 20)

Arguments

d	combined data matrix n*J, where n is number of subjects, J=J1+J2+... and J1 is number of features in omics dataset 1 and J2 is number of features in omics dataset 2...
K	number of clusters
gamma	Penalty on total number of features. Larger gamma will yeild small number of selected features.
alpha	balance between group sparsity and individual sparsity. alpha=1 yeilds no group sparsity. alpha=0 yeilds no individual penalty.
group	Prior group information. Potentially these group can contain overlap features. group is a list and each element of the list is feature index.
nstart	Number of initials for Kmeans for sparse Kmeans
wsPre	Initial feature weight.
penaltyInfo	only for the purpose of gap statitics. Here we will fix the penalty design to perform gap statistics. The input should be a list of groupInfo. See groupInfo for details.
sparseStart	Use Sparse Kmeans to do initialization.
silent	Output progress.
maxiter	Maximum numbre of iteration between ws and Cs.

Details

Integrative Sparse KMeans, integrating multiple omics dataset, using prior group information.

Value

m lists, m is length of gamma parameters. Each list is consisting of (ws=ws, Cs=Cs, objective=ADMMobject$objective, BIC=BIC, gamma=agamma,alpha=alpha

ws	weight for each feature. Zero weight means the feature is not selected.
Cs	Cluster Assignment
objective	objective value
obj0	sum of weighted separation ability. This term is for the purpose of gap statistics.
groupInfo	a list containing group design, alpha, gamma

Author(s)

Caleb

Examples

set.seed(123)

# Generate two random omics datasets
mu <- c(-3,1,3)
Simu1_mRNA <- rbind(cbind(matrix(rnorm(40*5, mu[1], 0.1),40,5),
                          matrix(rnorm(40*5, mu[2], 0.1),40,5),
                          matrix(rnorm(40*5, mu[3], 0.1),40,5)),
                    matrix(rnorm(10*15,0,0.1),10,15))

mu <- c(1,3,-3)
Simu1_methyl <- rbind(cbind(matrix(rnorm(40*5, mu[1], 0.1),40,5),
                            matrix(rnorm(40*5, mu[2], 0.1),40,5),
                            matrix(rnorm(40*5, mu[3], 0.1),40,5)),
                      matrix(rnorm(10*15,0,0.1),10,15))

## feature modules across two datasets
group <- list(c(1:10,51:60), c(11:20,61:70), c(21:30,71:80), c(31:40,81:90))

DList <- rbind(Simu1_mRNA, Simu1_methyl)
dim(DList)

## colSums(module)
K=3
nstart=20
silent=FALSE
maxiter=6
centers=NULL
error=1e-4
sparseStart = TRUE
gamma = 0.2
alpha = 0.5
d <- t(DList)

iRes <- ISKmeans(d, K=3, gamma=0.5, alpha=0.5, group=group)

Caleb-Huo/IS-Kmeans documentation built on May 6, 2019, 9:27 a.m.