meta.pca.cv: A algorithm of searching the optimal tuning parameter for...

In metaOmic/metaPCA: META-ANALYTIC PRINCIPAL COMPONENT ANALYSIS IN INTEGRATIVE OMICS APPLICATION

Usage

meta.pca.cv(DList, method, Meta.Dim, is.auto.Dim = TRUE, is.equal.Dim = FALSE, e.Dim, is.weight = TRUE, .var.quantile = 0.8, .scaleAdjust = FALSE, is.sparse = FALSE, CV_lambda = seq(1, 10, 1), is.plot = TRUE)

Arguments

DList	Input data set matrix (a list of multiple datasets; row=features, column=samples).
method	"SSC" = Sum of Squared Cosine, "SV" = Sum of variance
Meta.Dim	Dimension size of meta-eigenvector matrix
is.auto.Dim	Logical value whether dimension size of each study's eigenvector matrix (SSC) is determined by an arbitrary variance quantile
is.equal.Dim	Logical value whether dimension size of each study's eigenvector matrix (SSC) is equal across studies
e.Dim	Dimension size of each study's eigenvector matrix (SSC) when is.equal.Dim = TRUE
is.weight	Logical value whether the reciprocal of the largest eigenvalue is mutiplied to covariance matrix
.var.quantile	A threshold indicating the minimum variance of individual study, when is.auto.Dim = TRUE
.scaleAdjust	Logical value whether the PC projection is scaled to mean of zero and SD of 1
is.sparse	Logical value whether meta-eigenvector matrix is penalized to encourage sparseness
CV_lambda	A set of candidate tuning parameters in which the best tuning parameter is chosen
is.plot	Logical value whether visual scree plot is created

Value

The value of best tuning parameter selected among considered parameters

Author(s)

SungHwan Kim [e-mail]

Examples

library(yeastCC)
data(yeastCC)
data<-Biobase::exprs(yeastCC)
library(impute)
library(doMC)
data.na<-is.na(data)
data.na.length<-apply(data.na, 1, sum)
data.sd<-apply(as.matrix(data), 1, sd, na.rm=TRUE)

new.data<-data[data.na.length<77*0.1 & data.sd>0.45,]    
Spellman <- list(alpha=impute.knn(new.data[,5:22])$data,
                cdc15=impute.knn(new.data[,23:46])$data,
                cdc28=impute.knn(new.data[,47:63])$data,
                elu=impute.knn(new.data[,64:77])$data)
                
####################################################################################################
## Searching the optimal tuning parameter based on the proportion of increased explained variance
####################################################################################################

optimal.lambda <- meta.pca.cv(DList=Spellman, method="SSC", Meta.Dim=2, CV_lambda = seq(1,10,1), is.plot=TRUE)
## optimal.lambda = 8

metaOmic/metaPCA documentation built on May 22, 2019, 6:54 p.m.