188宝金博页面版

Stability ranking with the staRank packageJuliane SiebourgOctober 13, 2015Stability ranking can be used to obtain variable rankings that are highly reproducible. Such rankings areneeded for robust variable selection on univariate data. In most biological experiments several replicatemeasurement for a variable of interest, e.g. a gene, are available. An obvious way to combine andprioritize these values is to take their average or median. Other procedures like a t-test or rank sum testdeliver a statistic that also accounts for the variance in the data. Stability ranking provides a way ofranking measured elements based on one of the methods above and combining it with a bootstrappingroutine to estimate the stability with which an element occurs within the top k set of the ranking. Thefinal ranking orders the elements accoring to this stability. The theory behind the procedure is describedin [1]. Stability selection in general is described in [2]. This file contains two example stability analysisthat show how the staRank package should be used. In a first toy example, stability selection using allavailable base ranking methods is performed on simulated data. In a second example, an RNAi dataseton Salmonella infection [3] is analyzed.Simulation exampleFirst we create an artificial dataset of p genes (rows) with n replicate measurement values (columns).For each gene the effects are drawn from a normal distribution and their replicate variance is drawn froma gamma distribution.> # sample parameters for data> p<-20 # genes> n<-4 # replicates> trueEffects<-rnorm(p,0,5) # gene effect> s<-rgamma(p,shape=2.5,rate=0.5) # sample variance> # draw n replicate values of a gene> simData <- matrix(0,nr=p,nc=n,+ dimnames=list(paste("Gene",c(1:p)),paste("Replicate",c(1:n))))> for(i in 1:p){+ simData[i,]<-replicate(n,rnorm(1,mean=trueEffects[i],sd=s[i]))+ }Now stability ranking can be performed on the dataset, to find top scoring genes that are highly re-producible. Stability ranking can be applied using different base ranking methods. Implemented in thepackage are the mean, median and the test statistics of the rank sum test (Mann-Whitney test), t-testand RSA [4].> # load the stability ranking package> library(staRank)> # implemented ranking methods> method<-c('mean','median','mwtest','ttest','RSA')The ranking is performed calling the main function stabilityRanking. Here this is done using thedefault settings but for different base methods.> stabilityList<-list()> stabilityList[['mean']]<-stabilityRanking(simData,method='mean')1