tuneCluster.spca: Feature Selection Optimization for sPCA method

In timeOmics: Time-Course Multi-Omics data integration

Description

This function identify the number of feautures to keep per component and thus by cluster in mixOmics::spca by optimizing the silhouette coefficient, which assesses the quality of clustering.

Usage

tuneCluster.spca(X, ncomp = 2, test.keepX = rep(ncol(X), ncomp), ...)

Arguments

X	numeric matrix (or data.frame) with features in columns and samples in rows
ncomp	integer, number of component to include in the model
test.keepX	vector of integer containing the different value of keepX to test for block X.
...	other parameters to be included in the spls model (see mixOmics::spca)

Details

For each component and for each keepX value, a spls is done from these parameters. Then the clustering is performed and the silhouette coefficient is calculated for this clustering.

We then calculate "slopes" where keepX are the coordinates and the silhouette is the intensity. A z-score is assigned to each slope. We then identify the most significant slope which indicates a drop in the silhouette coefficient and thus a deterioration of the clustering.

Value

silhouette	silhouette coef. computed for every combinasion of keepX/keepY
ncomp	number of component included in the model
test.keepX	list of tested keepX
block	names of blocks
slopes	"slopes" computed from the silhouette coef. for each keepX and keepY, used to determine the best keepX and keepY
choice.keepX	best keepX for each component

Examples

demo <- suppressWarnings(get_demo_cluster())
X <- demo$X

# tuning
tune.spca.res <- tuneCluster.spca(X = X, ncomp = 2, test.keepX = c(2:10))
keepX <- tune.spca.res$choice.keepX
plot(tune.spca.res)

# final model
spca.res <- mixOmics::spca(X=X, ncomp = 2, keepX = keepX)
plotLong(spca.res)

timeOmics documentation built on Nov. 8, 2020, 10:58 p.m.