initializeZt: Initialization of the latent features of the c3co model
In pneuvial/c3co: Cancer Cell Clonality using Copy-Number

Description Usage Arguments Details Value References Examples

Initialization of the latent features of the c3co model

1
2
3

initializeZt(Y1, Y2 = NULL, K = min(dim(Y1)), flavor = c("hclust",
  "nmf", "archetypes", "svd", "subsampling"), stat = c("C1+C2", "C1",
  "C2"), verbose = FALSE)

`Y1`	A matrix containing the segmented minor copy number (n patients in rows and J segments in columns).
`Y2`	A matrix containing the segmented major copy number (n patients in rows and J segments in columns).
`K`	An integer value, the number of latent features in the model. Defaults to `min(dim(Y1))`.
`flavor`	A character value specifying how initialization is performed. Defaults to stats::hclust. See Details.
`stat`	Statistic used to perform initialization. Should be either `"C1+C2"`, `"C1"`, or `"C2"`.
`verbose`	A logical value indicating whether to print extra information. Defaults to `FALSE`.

The latent features are inferred as follows according to the value of argument flavor:

If flavor == "hclust" (default), the latent features are centers of clusters derived by hierarchical agglomerative clustering on the Euclidean distance between the input copy number profiles, and using Ward linkage (stats::hclust()).

If flavor == "nmf", the latent features are the coefficients of the non-negative matrix factorization in K of the input copy number profiles.

If flavor == "svd", the latent features are the first K right singular vectors of the singular value decomposition of the input copy number profiles. The flavor is not recommended as it may produce matrices with non-positive entries

If flavor == "archetypes", the latent features are defined using archetypal analysis.

If flavor == "subsampling", the latent features are chosen at random among existing profiles.

A list with two components:

Z1: An J-by-K matrix, the initial value for the J minor copy numbers of the K latent features
Z2: An J-by-K matrix, the initial value for the J major copy numbers of the K latent features

Warning: Note that the Z1 and Z2 components are actually the transposed version of those matrices. This notation mistake will be fixed in a future release.

Gaujoux R and Seoighe C (2010). A flexible R package for nonnegative matrix factorization. BMC Bioinformatics, 11(1), pp. 367.

Cutler A and Breiman L. (1994) Archetypal analysis. Technometrics, 36(4):338-3474.

## Simulate locus-level (C1,C2) copy-number data
dataAnnotTP <- acnr::loadCnRegionData(dataSet="GSE11976", tumorFrac=1)
dataAnnotN <- acnr::loadCnRegionData(dataSet="GSE11976", tumorFrac=0)
len <- 500*10  ## Number of loci
K <- 3L        ## Number of subclones
n <- 12L       ## Number of samples
bkps <- list(c(100, 250)*10, c(150, 400)*10, c(150, 400)*10)
regions <- list(c("(0,3)", "(0,1)", "(1,2)"),
                c("(1,1)", "(0,1)", "(1,1)"),
                c("(0,2)", "(0,1)", "(1,1)"))
datSubClone <- buildSubclones(len=len, nbClones=K, bkps=bkps, regions=regions,
                              dataAnnotTP=dataAnnotTP, dataAnnotN=dataAnnotN)
W <- rSparseWeightMatrix(nb.samp=n, nb.arch=K, sparse.coeff=0.90)
simu <- mixSubclones(subClones=datSubClone, W=W)

## Segment the copy-number data
seg <- segmentData(simu)

## Initialize C3CO model
Y1 <- t(seg$Y1)
Y2 <- t(seg$Y2)

resH <- initializeZt(Y1, Y2, K=K)  ## corresponds to flavor "hclust")
resNMF <- initializeZt(Y1, Y2, K=K, flavor="nmf")
## Not run: 
## often fails because of singularities:
resArch <- initializeZt(Y1, Y2, K=K, flavor="archetypes")

## End(Not run)
resSVD <- initializeZt(Y1, Y2, K=K, flavor="svd")
resC <- initializeZt(Y1, Y2, K=K, flavor="subsampling")

resNMF1 <- initializeZt(Y1, K=K, flavor="nmf")