An R package for performing top-scoring pairs regression. The basic idea is to look for pairs of genes that switch rankings between phenotypes. The tspreg package allows the discovery of top-scoring pairs for multi-class classification and for survival classification.
## If needed
install.packages("devtools")
## Install pre-requisite packages from Bioconductor
source("http://bioconductor.org/biocLite.R")
biocLite("genefilter")
source("http://bioconductor.org/biocLite.R")
biocLite("tspair")
## Install the survhd package from Bitbucket
install_bitbucket(repo="survhd",user="lwaldron",ref="default")
library(tspair)
library(tspreg)
data(tspdata)
## Build a top-scoring pairs regression predictor
tspr1 <- tspreg(dat,as.factor(grp),npair=1)
# Shows the pairs of indices contributing to the predictor
tspr1$index
# Create the data frame for prediction
predictors <- calculateTspairs(dat,tspr1$index)
# Fit the linear model and predict
modelFit <- lm(grp=="diseased" ~ .,data=predictors$pairMat)
# Get the pairs on the test data set
validationPredictors <- calculateTspairs(dat2,tspr1$index)
# Apply predictor to test data set
predict(modelFit,newdata=validationPredictors$pairMat) > 0.5
Please use the following to find citations for the tspreg package citation("tspreg")
.