permutationMultipleLmMatrix {GIGSEA} | R Documentation |
permutationMultipleLmMatrix is a permutation test to calculate the empirical p values for weighted multiple linear regression
permutationMultipleLmMatrix(fc, net, weights = rep(1, nrow(net)), num = 100, step = 1000, verbose = TRUE)
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 |
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.
Shijia Zhu, shijia.zhu@mssm.edu
orderedIntersect
;
permutationMultipleLm
;
# 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, by.x=data$gene, by.y=rownames(net)) net2 <- orderedIntersect( x=net, by.x=rownames(net), by.y=data$gene) all( rownames(net2) == as.character(data2$gene) ) # the MGSEA.res1 uses the weighted multiple linear regression to do # permutation test, # while MGSEA.res2 used the solution of weighted matrix operation. The # latter one takes substantially less time. # system.time( MGSEA.res1<-permutationMultipleLm(fc=data2$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)