Mercurial > repos > jasonxu > matrixeqtl
view MatrixEQTL/demo/b.cvrt.r @ 0:cd4c8e4a4b5b draft
Uploaded
author | jasonxu |
---|---|
date | Fri, 12 Mar 2021 08:12:46 +0000 |
parents | |
children |
line wrap: on
line source
library("MatrixEQTL"); # Number of columns (samples) n = 100; # Number of covariates nc = 10; # Generate the covariates cvrt.mat = 2 + matrix(rnorm(n*nc), ncol = nc); # Generate the vectors with genotype and expression variables snps.mat = cvrt.mat %*% rnorm(nc) + rnorm(n); gene.mat = cvrt.mat %*% rnorm(nc) + rnorm(n) + 0.5 * snps.mat + 1; # Create 3 SlicedData objects for the analysis snps1 = SlicedData$new( matrix( snps.mat, nrow = 1 ) ); gene1 = SlicedData$new( matrix( gene.mat, nrow = 1 ) ); cvrt1 = SlicedData$new( t(cvrt.mat) ); # Produce no output files filename = NULL; # tempfile() # Call the main analysis function me = Matrix_eQTL_main( snps = snps1, gene = gene1, cvrt = cvrt1, output_file_name = filename, pvOutputThreshold = 1, useModel = modelLINEAR, errorCovariance = numeric(), verbose = TRUE, pvalue.hist = FALSE ); # Pull Matrix eQTL results - t-statistic and p-value beta = me$all$eqtls$beta; tstat = me$all$eqtls$statistic; pvalue = me$all$eqtls$pvalue; rez = c(beta = beta, tstat = tstat, pvalue = pvalue); # And compare to those from the linear regression in R { cat("\n\n Matrix eQTL: \n"); print(rez); cat("\n R summary(lm()) output: \n"); lmdl = lm( gene.mat ~ snps.mat + cvrt.mat ); lmout = summary(lmdl)$coefficients[2,c("Estimate","t value","Pr(>|t|)")]; print( lmout ); } # Results from Matrix eQTL and "lm" must agree stopifnot(all.equal(lmout, rez, check.attributes=FALSE));