MISKmeans: MISKmeans
In Caleb-Huo/MIS-Kmeans:

Description Usage Arguments Details Value Author(s) Examples

MISKmenas

MISKmeans(d, K = NULL, gamma = NULL, lambda = 0.5, alpha = 0.5,
  group = NULL, nstart = 20, wsPre = NULL, iniWbound = 20,
  penaltyInfo = NULL, silent = FALSE, maxiter = 20,
  sampleSizeAdjust = FALSE)

`d`	A list of S studies, each study is a 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.
`lambda`	A tuning parameter controlling the balance between separation ability (BCSS/TSS) and matching function. lambda is set to be 0.5 by default.
`alpha`	balance between group sparsity and individual sparsity. alpha=1 yeilds no group sparsity. alpha=0 yeilds no individual penalty. Default alpha=0.5
`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.
`silent`	Output progress.
`maxiter`	Maximum numbre of iteration between ws and Cs.
`sampleSizeAdjust`	logical argument, controlling whether to adjust for sample size. If true, that means study with larger sample size will have a larger impact. If false, each study has equal contribution. Without prior information, sampleSizeAdjust=FALSE is suggested since we are not sure about data quality.

Meta-analytic multi-level omics data integration with incorporation of prior group structure

m lists, m is length of gamma parameters. If gamma is a scalar, m = 1. The following items are included in the list.

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

Caleb

S <- 2
K <- 3
G <- 1000
g1 <- 50
g2 <- 50
n0 <- 20
n <- K*n0
labels <- cut(1:n,breaks=K,labels=FALSE)

set.seed(32611)
S1 <- matrix(rnorm(G*n), nrow=G, ncol=n)
S2 <- matrix(rnorm(G*n), nrow=G, ncol=n)

S1[1:g1, labels==1] <- S1[1:g1, labels==1] + 2
S1[1:g1, labels==3] <- S1[1:g1, labels==3] - 2
S1[g1 + 1:g2, labels==1] <- S1[g1 + 1:g2, labels==1] - 2
S1[g1 + 1:g2, labels==2] <- S1[g1 + 1:g2, labels==2] + 2

heatmap(S1, Rowv=NA, Colv=NA)

S2[1:g1, labels==2] <- S2[1:g1, labels==2] + 2
S2[1:g1, labels==1] <- S2[1:g1, labels==1] - 2
S2[g1 + 1:g2, labels==2] <- S2[g1 + 1:g2, labels==2] - 2
S2[g1 + 1:g2, labels==3] <- S2[g1 + 1:g2, labels==3] + 2

heatmap(S2, Rowv=NA, Colv=NA)

S = list(t(S1),t(S2))
groups <- Map('c',1:g1,g1 + 1:g2)

res <- MISKmeans(d = S, K = 3, gamma = 0.4, group = groups)

names(res)

order_S1 <- order(res$Cs[[1]])
order_S2 <- order(res$Cs[[2]])

S1_sub <- S1[res$ws!=0, order_S1]
S2_sub <- S2[res$ws!=0, order_S2]

col_S1 <- palette()[res$Cs[[1]][order_S1]]
col_S2 <- palette()[res$Cs[[2]][order_S2]]

heatmap(S1_sub, Rowv=NA, Colv=NA, ColSideColors=col_S1)
heatmap(S2_sub, Rowv=NA, Colv=NA, ColSideColors=col_S2)