permutationMultipleLmMatrix: permutationMultipleLmMatrix

In zhushijia/GIGSEA: Genotype Imputed Gene Set Enrichment Analysis

Description

permutationMultipleLmMatrix is a permutation test to calculate the empirical p values for weighted multiple linear regression

Usage

permutationMultipleLmMatrix(fc, net, weights = rep(1, nrow(net)),
  num = 100, step = 1000, verbose = TRUE)

Arguments

fc	a vector of numeric values representing gene expression fold change
net	a matrix of numeric values in the size of gene number x gene set number, representing the connectivity between genes and gene sets
weights	a vector of numeric values representing the weights of permuated genes
num	an integer value representing the number of permutations
step	an integer value representing the number of permutations in each step
verbose	an boolean value indicating whether or not to print output to the screen

Value

a data frame comprising following columns:

• term a vector of character values incidating the names of gene sets.

• usedGenes a vector of numeric values indicating the number of genes used in the model.

• observedTstats a vector of numeric values indicating the observed t-statistics for the weighted multiple regression coefficients.

• empiricalPval a vector of numeric values [0,1] indicating the permutation-based empirical p values.

• BayesFactor a vector of numeric values indicating the Bayes Factor for the multiple test correction.

Author(s)

Shijia Zhu, [e-mail]

See Also

orderedIntersect; permutationMultipleLm;

Examples

# load data
data(heart.metaXcan)
gene <- heart.metaXcan$gene_name

# extract the imputed Z-score of differential gene expression, which 
# follows the normal distribution
fc <- heart.metaXcan$zscore

# use as weights the prediction R^2 and the fraction of imputation-used SNPs 
usedFrac <- heart.metaXcan$n_snps_used / heart.metaXcan$n_snps_in_cov
r2 <- heart.metaXcan$pred_perf_r2
weights <- usedFrac*r2

# build a new data frame for the following weighted linear regression-based 
# enrichment analysis
data <- data.frame(gene,fc,weights)
head(data)

net <- MSigDB.KEGG.Pathway$net

# intersect the imputed genes with the gene sets of interest
data2 <- orderedIntersect( x = data[,c("fc","weights")] , by.x = data$gene , 
by.y = rownames(net)  )
net2 <- orderedIntersect( x = net , by.x = rownames(net) , 
by.y = data$gene  )
all( rownames(net2) == rownames(data2) )

# the MGSEA.res1 uses the weighted multiple linear regression to do 
# permutation test, 
# while MGSEA.res2 used the solution of weighted matrix operation. The 
# latter one takes substantially less time.
# system.time( MGSEA.res1<-permutationMultipleLm(fc=data2$fc, net=net2, 
# weights=data2$weights, num=1000))
system.time( MGSEA.res2<-permutationMultipleLmMatrix(fc=data2$fc, net=net2, 
weights=data2$weights, num=1000))
head(MGSEA.res2)

zhushijia/GIGSEA documentation built on May 3, 2019, 10:45 p.m.