tuningParametersCombROC: Test different models using different kernel of SVM, values...

In guidmt/SVM2CRM: SVM2CRM: support vector machine for cis-regulatory elements detections

Description

The function tuningParametersCombROC allow high flexibility: the user can set the type of kernel, the cost parameter of SVM, the number of histone marks. tuningParametersCombROC use performanceSVM and the output is a data.frame where for each model there are the parameters compute with performanceSVM. To help the user to discriminate how to discriminate the best model SVM2CRM implement several function to plot the results of performanceSVM.

Usage

 tuningParametersCombROC(training_set, typeSVM, costV,different.weight=TRUE, vecS,infofile,pcClass.string="enhancers",nClass.string="not_enhancers",pcClass,ncClass) 

Arguments

training_set	training set (data.frame)
typeSVM	a vector with the kinds of Kernel
costV	a vector with a list of cost parameters
different.weight	the data are unbalanced (default TRUE)
vecS	a vector with the number of histone marks (e.g. from 2 to x)
infofile	a data.frame where in the column "a" there are the histone marks, while in the column "b" a vectors of letters.
pcClass.string	label of the first class (e.g. "enhancers")
nClass.string	label of the second class (e.g. "not_enhancers")
pcClass	number of positive class in the training set
ncClass	number of negative class in the training set

Details

Some detailled description

Value

A data.frame with the values from perfomanceSVM for each trained model.

Author(s)

Guidantonio Malagoli Tagliazucchi [e-mail]

See Also

cisREfindbed, perfomanceSVM, plotFscore, plotROC

Examples

   library("GenomicRanges")
   library("SVM2CRMdata")

   setwd(system.file("data",package="SVM2CRMdata"))
   load("CD4_matrixInputSVMbin100window1000.rda")
   completeTABLE<-CD4_matrixInputSVMbin100window1000

   new.strings<-gsub(x=colnames(completeTABLE[,c(6:ncol(completeTABLE))]),pattern="CD4.",replacement="")
   new.strings<-gsub(new.strings,pattern=".norm.w100.bed",replacement="")
   colnames(completeTABLE)[c(6:ncol(completeTABLE))]<-new.strings

   #Create a vector that contain the list of the bed files that you want use during the analysis
   #list_file<-grep(dir(),pattern=".sort.txt",value=TRUE)
   #print(list_file)
   #Here we used a data.frame that contain the genomic coordinates of p300 binding sites
   #train_positive<-getSignal(list_file,chr="chr1",reference="p300.distal.fromTSS.txt",win.size=500,bin.size=100,label1="enhancers")
   #train_negative<-getSignal(list_file,chr="chr1",reference="random.region.hg18.nop300.txt",win.size=500,bin.size=100,label1="not_enhancers")
   setwd(system.file("data",package="SVM2CRMdata"))
   load("train_positive.rda")
   load("train_negative.rda")

   training_set<-rbind(train_positive,train_negative)
   colnames(training_set)[c(5:ncol(training_set))]<-gsub(x=gsub(x=colnames(training_set[,c(5:ncol(training_set))]),pattern="sort.txt.",replacement=""),pattern="CD4.",replacement="")


   setwd(system.file("extdata", package = "SVM2CRMdata"))
   data_level2 <- read.table(file = "GSM393946.distal.p300fromTSS.txt",sep = "\t", stringsAsFactors = FALSE)
   data_level2<-data_level2[data_level2[,1]=="chr1",]

   DB <- data_level2[, c(1:3)]
   colnames(DB)<-c("chromosome","start","end")

   label <- "p300"

   table.final.overlap<-findFeatureOverlap(query=completeTABLE,subject=DB,select="all")

   data_enhancer_svm<-createSVMinput(inputpos=table.final.overlap,inputfull=completeTABLE,label1="enhancers",label2="not_enhancers")
   colnames(data_enhancer_svm)[c(5:ncol(data_enhancer_svm))]<-gsub(gsub(x=colnames(data_enhancer_svm[,c(5:ncol(data_enhancer_svm))]),pattern="CD4.",replacement=""),pattern=".norm.w100.bed",replacement="")

   listcolnames<-c("H2AK5ac","H2AK9ac","H3K23ac","H3K27ac","H3K27me3","H3K4me1","H3K4me3")

   dftotann<-smoothInputFS(train_positive[,c(6:ncol(train_positive))],listcolnames,k=20)

   results<-featSelectionWithKmeans(dftotann,5)

   resultsFS<-results[[7]]

   resultsFSfilter<-resultsFS[which(resultsFS[,2]>median(resultsFS[,2])),]

   resultsFSfilterICRR<-resultsFSfilter[which(resultsFSfilter[,3]<0.50),]

   listHM<-resultsFSfilterICRR[,1]
   listHM<-gsub(gsub(listHM,pattern="_.",replacement=""),pattern="CD4.",replacement="")


   selectFeature<-grep(x=colnames(training_set[,c(6:ncol(training_set))]),pattern=paste(listHM,collapse="|"),value=TRUE)

   colSelect<-c("chromosome","start","end","label",selectFeature)
   training_set<-training_set[,colSelect]


   vecS <- c(2:length(listHM))
   typeSVM <- c(0, 6, 7)[1]
   costV <- c(0.001, 0.01, 0.1, 1, 10, 100, 1000)[6]
   wlabel <- c("not_enhancer", "enhancer")
   infofile<-data.frame(a=c(paste(listHM,"signal",sep=".")))
   infofile[,1]<-gsub(gsub(x=infofile[,1],pattern="CD4.",replacement=""),pattern=".sort.bed",replacement="")


   tuningTAB <- tuningParametersCombROC(training_set = training_set, typeSVM = typeSVM, costV = costV,different.weight="TRUE", vecS = vecS[1],pcClass=100,ncClass=400,infofile)

guidmt/SVM2CRM documentation built on Dec. 20, 2021, 1:48 p.m.