library(tibble)
library(dplyr)
library(bmrm)
library(pROC)
library(glmnet)
library(xlsx)
library(dplyr)
#Functions
writeCSVMultiple <- function(x,file){
if(dir.exists("file")!=F){
cat("Creating folder",file,"\n")
dir.create(file)
}
for(i in seq_along(x)){
write.csv(x = x[[i]],
file = paste(file, "_",names(x)[[i]],".csv",sep = ""))
}
}
QFunclog <- function(x,Threshold)
ifelse(x>Threshold,log10(x+1),0)
QFuncjumpramp <- function(x,Threshold)
ifelse(x>Threshold,x,0)
printRemoveDropResults <- function(ThisResult,whichRound="Round 4"){
cat("Sensitivity:",(mean(unlist(ThisResult$AllSensitivity))),"\n")
print("Only this protein:")
print((rownames(ThisResult$AllProbPredictionsMinusFeatureNIter) %>%
sapply(FUN = function(x) ThisResult$AllProbPredictionsMinusFeatureNIter[[x,"LR"]]%>%
dplyr::select(-BV) %>%
summarise_all(.funs = funs(mean(.[which(CancerStatus==T)]>ThisResult$AllThresh99NIter[[x,"LR"]]))))) %>%
apply(MARGIN = c(1,2),as.numeric) %>%
rowMeans()%>% sort()%>% t()%>% t())
print("Remove this protein:")
print((rownames(ThisResult$AllProbPredictionsMinusFeatureNIter) %>%
sapply(FUN = function(x) ThisResult$AllProbPredictionsDropFeatureNIter[[x,"LR"]]%>%
dplyr::select(-BV) %>%
summarise_all(.funs = funs(mean(.[which(CancerStatus==T)]>ThisResult$AllThresh99NIter[[x,"LR"]]))))) %>%
apply(MARGIN = c(1,2),as.numeric) %>%
rowMeans()%>%sort()%>%t()%>%t())
mycoefs <- ThisResult$Alllogit %>% coef(s=0) %>% abs()
FirstIteration <- ThisResult$Master4ClassificationwithClassAll[whichRound,"LR"][[1]]
MinDiff <- FirstIteration %>%
mutate_at(.vars = setdiff(colnames(FirstIteration),c("CancerStatus","maxOmega","fold")),
.funs=funs(QFuncjumpramp(.,quantile(.[which(CancerStatus==T)],probs=0.95)))) %>%
summarise_at(.vars = setdiff(colnames(FirstIteration),
c("CancerStatus","fold")),
.funs = funs(mean(.[which(CancerStatus==T)])-mean(.[which(CancerStatus==F)]))) %>%
as.data.frame()
RankingCoefs <- data.frame('Ranking Score'= as.vector(MinDiff*mycoefs[names(MinDiff),1]) %>%
t(),
'Coefs' = mycoefs[names(MinDiff),1] %>%
t()
%>%t())
print(RankingCoefs%>%
rownames_to_column("Protein")%>%
arrange(Ranking.Score))
}
ProbsEachRound <- function(ThisResult,csvFile=NULL){
Probs = ThisResult$AllProbPredictionsNIter[,"LR"] %>% sapply(FUN = function(x) x)
Probs <- rbind(Probs,Thresh=ThisResult$AllThresh99NIter[,"LR"] %>% sapply(FUN = function(x) x))
Probs <- cbind(Probs,
CancerStatus =
ThisResult$Master4ClassificationwithClassAll[["Round 1","LR"]][rownames(Probs),"CancerStatus"])
if(!is.null(csvFile))
write.csv(file = csvFile,x = Probs)
return(Probs)
}
Generate4LudaObject <- function(ThisResult,rdaFile,TissueType=TissueType_RFData){
ForLuda <- sapply(setNames(paste("Round",1:10),paste("Round",1:10)),
function(x) cbind(ThisResult$Master4ClassificationwithClassAll[[x,"LR"]],
LogitCall=ThisResult$AllpredictionsNIter[[x,"LR"]],
LogitProb=ThisResult$AllProbPredictionsNIter[[x,"LR"]],
Tissue=TissueType_RFData[rownames(ThisResult$Master4ClassificationwithClassAll[[x,"LR"]])
]),simplify = F)
ForLudaFeatureRanks <- ThisResult$AlllogitFeaturesRank
ForLudaSensitivity <- ThisResult$AllSensitivity
save(file=rdaFile,
list= c("ForLuda","ForLudaFeatureRanks","ForLudaSensitivity","ThisResult"))
return( list(ForLuda=ForLuda,
ForLudaFeatureRanks=ForLudaFeatureRanks,
ForLudaSensitivity=ForLudaSensitivity,
ThisResult=ThisResult))
}
cv.CancervsNormal.QuantileReplace.Mutatation.knownCV <- function(CancerStatus,
# a vector containing cancer status (T=Cancer)
Proteins,
# a matrix of protien vector
Mutations,
# a matrix of Mutations
# It must at include BV.,
# iteration , fold
#Methods=c("RF","Logit"),
# Methods to test
ProteinsHigherCancer = T,
#If true, we only keep proteins with higher median in cancer in the sense of wilcoxon
ForcePositiveCoeff = F,
#If true, we force the coefficients to be positive (This is togher measure than ProteinsHigherCancer)
#myseed=1234,
#NIter=10,
#nfold=10,
SpecSet = 0.99, #Specificity we shoot for
FeaturesToQuantile, # Features to quantile, set to Null if you do not wish quantiling
QuantileF=0.95,
QuantileFunction= function(x,Threshold)
ifelse(x>Threshold,x,0),
NoPenaltyonCtdna=F,
NoLassoPenaly = F){
mylower <- ifelse(ForcePositiveCoeff,0,-Inf)
if(ProteinsHigherCancer){
#apply Wilcoxon to filter proteins if ProteinsHigherCancer==T
Proteins2Use <- (apply(Proteins,MARGIN = 2,
FUN = function(x) wilcox.test(
x[which(CancerStatus==T)],
x[which(CancerStatus==F)],
alternative = "greater")$p.value)<0.05) %>%
which() %>%
names()
# Proteins2Use <- (colMeans(Proteins[which(CancerStatus==T),,drop=F])-
# colMeans(Proteins[which(CancerStatus==F),,drop=F])>0) %>%
# which() %>%
# names()
}else{
Proteins2Use <- colnames(Proteins)
}
# if(is.null(Mutations)){
# Master4ClassificationwithClass <- data.frame(CancerStatus=CancerStatus,
# Proteins[,Proteins2Use,drop=F])
# }else{
#}
#Remove the penalty for ctdna if NoPenaltyonCtdna==T
if(NoPenaltyonCtdna){
PenaltyFactor <- c(rep(1,length(Proteins2Use)),rep(0,ncol(Mutations)-3))
}else{
PenaltyFactor <-c(rep(1,length(Proteins2Use)),rep(1,ncol(Mutations)-3))
}
#Only FeaturesToQuantile which have survived the Wilcoxon are being quantiles
if(!is.null(FeaturesToQuantile))
FeaturesToQuantile <- intersect(FeaturesToQuantile,Proteins2Use)
#RFFeatures <- colnames(Master4ClassificationwithClass) %>% setdiff("CancerStatus")
#augmenting the quantiled and non-quantiled proteins
MutationsRound1st <- Mutations %>%
filter(iteration==1) %>%
column_to_rownames("BV")
MutationsRound1st <- MutationsRound1st[rownames(Proteins),]
MutationsRound1st[is.na(MutationsRound1st)] <- 0
Master4ClassificationwithClass <- data.frame(CancerStatus=CancerStatus,
Proteins[,Proteins2Use,drop=F],
MutationsRound1st %>%
dplyr::select(-iteration,-fold))
if(is.null(FeaturesToQuantile)){
Master4ClassificationwithClassAppQuantAug <-
Master4ClassificationwithClass
}else{
ThrehsoldsTraining <- Master4ClassificationwithClass %>%
filter(CancerStatus==F) %>%
summarise_at(.vars = FeaturesToQuantile,# to implement ramp you can manipulate here
.funs = funs(quantile(.,probs = QuantileF)))
Master4ClassificationwithClassAppQuantAug <- # to implement ramp you can manipulate here
Master4ClassificationwithClass %>%
mutate_(.dots = setNames(paste("QuantileFunction(",
names(ThrehsoldsTraining),
",",
ThrehsoldsTraining,")",
sep = ""),
names(ThrehsoldsTraining)))
}
#Do cross-validation without quantiling for the Logit
Alllogit <- cv.glmnet(x= Master4ClassificationwithClassAppQuantAug %>%
dplyr::select(-CancerStatus) %>%
as.matrix(),
y=Master4ClassificationwithClass$CancerStatus,
type.measure = "auc",
family="binomial",
nfolds = length(unique(MutationsRound1st$fold)),
penalty.factor = PenaltyFactor,
lower.limits=mylower,
keep=TRUE)
AllLogitThreshod <- apply(Alllogit$fit.preval[which(Master4ClassificationwithClass$CancerStatus==F),],
MARGIN = 2,
FUN = quantile,probs=SpecSet,na.rm=T) %>%
as.numeric() ## to implement ramp you can manipulate here
AllSensity <- sapply(seq_along(AllLogitThreshod),
FUN = function(x)
mean(Alllogit$fit.preval[ which(Master4ClassificationwithClass$CancerStatus==T),x]
>AllLogitThreshod[x]))
#find the best lambda suited for desired SpecSet
if(NoLassoPenaly){
lambda.best <- 0
}else{
lambda.best <- Alllogit$lambda[which.max(AllSensity)]
}
AlllogitFeatures <- which(as.matrix(coef(Alllogit,s=lambda.best))!=0,
arr.ind = T,
useNames = T) %>%
rownames() %>%
setdiff("(Intercept)")
NIter <- unique(Mutations$iteration)
AllProbPredictionsNIter <- matrix(data = list(),
ncol = 1,
nrow = length(NIter),
dimnames = list(paste("Round",NIter),
"LR"))
AllThresh99NIter <- AllProbPredictionsNIter
AllpredictionsNIter <- AllProbPredictionsNIter
AllProbPredictionsNIter <- AllProbPredictionsNIter
AllSensitivity <- AllProbPredictionsNIter
Master4ClassificationwithClassAll <- AllProbPredictionsNIter
AllProbPredictionsMinusFeatureNIter <- AllProbPredictionsNIter
AllProbPredictionsDropFeatureNIter <- AllProbPredictionsNIter
#set.seed(myseed)
#AllMyIndices <- sapply(1:NIter,FUN = function(x)
# balanced.cv.fold(Master4ClassificationwithClass$CancerStatus) %>% as.numeric,simplify = F)
#names(AllMyIndices)<- paste("Round",NIter,sep = "")
set.seed(1234)
for( j in NIter){
#MyIndices <- AllMyIndices[[j]]
ThisRound <- Mutations %>%
filter(iteration==j)
uniqueMyIndices <- ThisRound %>%
dplyr::select(fold) %>%
unlist() %>%
unique()
Allpredictions <- setNames(vector(length = nrow(Proteins)),rownames(Proteins))
ProbPredictions <- setNames(vector(length = nrow(Proteins),mode = "numeric") ,rownames(Proteins))
ProbPredictionsMinusFeature <- matrix(0,
nrow=nrow(Proteins),
ncol= ncol(Master4ClassificationwithClass)-1 ,
dimnames = list(rownames(Proteins),Master4ClassificationwithClass %>% colnames() %>% setdiff("CancerStatus"))
)
ProbPredictionsDropFeature <- ProbPredictionsMinusFeature
#for(k in seq_along(Methods)){
MutationsThisRound <- ThisRound %>%
column_to_rownames("BV")
MutationsThisRound <- MutationsThisRound[rownames(Proteins),]
MutationsThisRound[is.na(MutationsThisRound)] <- 0
Master4ClassificationwithClass <- data.frame(CancerStatus=CancerStatus,
Proteins[,Proteins2Use,drop=F],
MutationsThisRound %>%
dplyr::select(-iteration,-fold))
for( i in uniqueMyIndices){
trainSamples <- ThisRound %>% #which(MyIndices!=i)
filter(fold!=i) %>%
dplyr::select(BV) %>%
unlist()
testSamples <- ThisRound %>% #which(MyIndices!=i)
filter(fold==i) %>%
dplyr::select(BV) %>%
unlist()
if(is.null(FeaturesToQuantile)){
Master4ClassificationwithClassAppQuantAug <-
Master4ClassificationwithClass
}else{
ThrehsoldsTraining <- Master4ClassificationwithClass[trainSamples,] %>%
filter(CancerStatus==F) %>%
summarise_at(.vars = FeaturesToQuantile,# to implement ramp you can manipulate here
.funs = funs(quantile(.,probs = QuantileF)))
Master4ClassificationwithClassAppQuantAug <- # to implement ramp you can manipulate here
Master4ClassificationwithClass %>%
mutate_(.dots = setNames(paste("QuantileFunction(",
names(ThrehsoldsTraining),
",",
ThrehsoldsTraining,")",
sep = ""),
names(ThrehsoldsTraining)))
rownames(Master4ClassificationwithClassAppQuantAug) <- rownames(Master4ClassificationwithClass)
}
#if(Methods[k]=="Logit"){
thislogit <- glmnet(x= Master4ClassificationwithClassAppQuantAug[trainSamples,] %>%
dplyr::select(-CancerStatus) %>%
as.matrix(),
y=Master4ClassificationwithClassAppQuantAug[trainSamples,"CancerStatus"],
#lambda = lambda.best,
penalty.factor = PenaltyFactor,
family="binomial",
lower.limits=mylower)
ProbPredictions[testSamples] <- predict(thislogit,
Master4ClassificationwithClassAppQuantAug[testSamples,]%>%
dplyr::select(-CancerStatus) %>%
as.matrix(),
type="response",
s=lambda.best) %>%
unlist()
for( k in Master4ClassificationwithClassAppQuantAug%>% colnames() %>% setdiff("CancerStatus")){
z <- Master4ClassificationwithClassAppQuantAug[testSamples,] %>%
dplyr::select(-CancerStatus) %>%
as.matrix()
z[,setdiff(colnames(z),k)] <- 0
ProbPredictionsMinusFeature[testSamples,k]<- predict(thislogit,
z,
type="response",
s=lambda.best) %>%
unlist()
z <- Master4ClassificationwithClassAppQuantAug[testSamples,] %>%
dplyr::select(-CancerStatus) %>%
as.matrix()
z[,k] <- 0
ProbPredictionsDropFeature[testSamples,k]<- predict(thislogit,
z,
type="response",
s=lambda.best) %>%
unlist()
}
#}else if(Methods[k]=="RF"){
# ThisRF <- randomForest(as.factor(CancerStatus)~.,
# data =Master4ClassificationwithClassAppQuantAug[,c("CancerStatus",RFFeatures)] ,
# subset=trainSamples,
# cutoff = c(0.15,0.85))
# ProbPredictions[testSamples] <- predict(ThisRF,Master4ClassificationwithClassAppQuantAug[testSamples,],type="prob")[,"TRUE"]
#}
}
ThisThresh99 <- ProbPredictions[which(Master4ClassificationwithClass$CancerStatus==F)] %>%
quantile(SpecSet)
Allpredictions <- ProbPredictions>ThisThresh99
AllThresh99NIter[[j]] <- ThisThresh99
AllpredictionsNIter[[j]] <- Allpredictions
AllProbPredictionsNIter[[j]] <- ProbPredictions
AllProbPredictionsMinusFeatureNIter[[j]] <- as.data.frame(ProbPredictionsMinusFeature) %>%
rownames_to_column("BV") %>%
mutate( CancerStatus =
Master4ClassificationwithClass[rownames(ProbPredictionsMinusFeature),
"CancerStatus"])
AllProbPredictionsDropFeatureNIter[[j]] <- as.data.frame(ProbPredictionsDropFeature) %>%
rownames_to_column("BV") %>%
mutate( CancerStatus =
Master4ClassificationwithClass[rownames(ProbPredictionsMinusFeature),
"CancerStatus"])
AllSensitivity[[j]] <- Allpredictions[which(Master4ClassificationwithClass$CancerStatus==T)] %>%
mean()
Master4ClassificationwithClassAll[[j]] <- cbind(Master4ClassificationwithClass,
fold=
MutationsThisRound[rownames(Master4ClassificationwithClass),"fold"])
#}
}
return(list(Master4ClassificationwithClassAll=Master4ClassificationwithClassAll,
AllThresh99NIter=AllThresh99NIter,
AllpredictionsNIter=AllpredictionsNIter,
AllProbPredictionsNIter=AllProbPredictionsNIter,
AllProbPredictionsMinusFeatureNIter =AllProbPredictionsMinusFeatureNIter,
AllProbPredictionsDropFeatureNIter = AllProbPredictionsDropFeatureNIter,
AllSensitivity=AllSensitivity,
AllSensitivitycvglmnet=max(AllSensity,na.rm = T), # This comes out of Logit cv and more reliable
AllcvglmnetCalls = Alllogit$fit.preval[,which.max(AllSensity)],
ProteinsHigherCancer=ProteinsHigherCancer,
NoMutations=is.null(Mutations),
Proteins2Use = Proteins2Use,
Alllogit=Alllogit,
AlllogitFeatures=AlllogitFeatures,
#RFFeatures=RFFeatures,
lambda.best=lambda.best))
}
#Preparing Data
Master_Protein_Dataset_12_9_17_Logist <- read.xlsx2("MEGA Protein Dataset for Logistic Regression, December 9, 2017.xlsx",
sheetIndex = 1,
stringsAsFactors = F)
Master_Protein_Dataset_12_9_17_RF <- read.xlsx2("MEGA Protein Dataset for Random Forest, December 9, 2017.xlsx",
sheetIndex = 1,
stringsAsFactors = F)
Master_Protein_Filtered_RF <-
#Master_Protein_Dataset_20_10_17 %>%
Master_Protein_Dataset_12_9_17_RF %>%
#filter(!(BV. %in% PostSample)) %>%
#Master_Protein_Dataset_9_16_17 %>%
#filter(BV. %in% Master_Protein_Dataset_1_10_17$BV.)%>%
dplyr::select(-`LRG.1`,-Vitronectin)
TissueType_RFData <- setNames(Master_Protein_Filtered_RF$Sample.Type,Master_Protein_Filtered_RF$BV.)
# ajdusting for limits of detection for each protein?
OnlyProteins_RF <- Master_Protein_Filtered_RF %>%
dplyr::select(-BV.,-Stage,-Sample.Type) %>%
apply(MARGIN = 2,FUN = as.numeric) %>%
as.data.frame()
OnlyProteins_RF[is.na(OnlyProteins_RF)] <- 0
colnames(OnlyProteins_RF) <- gsub(colnames(OnlyProteins_RF),pattern = "[.]",replacement = "")
rownames(OnlyProteins_RF) <- Master_Protein_Filtered_RF$BV.
Master_Protein_Filtered_LR <-
#Master_Protein_Dataset_20_10_17 %>%
Master_Protein_Dataset_12_9_17_Logist %>%
#filter(!(BV. %in% PostSample)) %>%
#Master_Protein_Dataset_9_16_17 %>%
#filter(BV. %in% Master_Protein_Dataset_1_10_17$BV.)%>%
dplyr::select(-`LRG.1`,-Vitronectin)
# ajdusting for limits of detection for each protein?
OnlyProteins_LR <- Master_Protein_Filtered_LR %>%
dplyr::select(-BV.,-Stage,-Sample.Type) %>%
apply(MARGIN = 2,FUN = as.numeric) %>%
as.data.frame()
OnlyProteins_LR[is.na(OnlyProteins_LR)] <- 0
colnames(OnlyProteins_LR) <- gsub(colnames(OnlyProteins_LR),pattern = "[.]",replacement = "")
rownames(OnlyProteins_LR) <- Master_Protein_Filtered_LR$BV.
## CtdNA
load("maxValuesPerSample_20171209_FORJosh.rda")
sampleTable_ALLModified <-
sampleTable_ALL[,c("iteration","fold","maxOmega","CosmicCount","Sample.Category")] %>%
rownames_to_column("BV") %>%
mutate(BV=ifelse(iteration==1,BV,substr(BV,1,nchar(BV)-1)))
NewCtDNA_ALLRounds <- sampleTable_ALLModified %>%
filter(BV %in% Master_Protein_Filtered_LR$BV.)%>%
transmute(BV=BV,
iteration=iteration,
fold=fold,
maxOmega= maxOmega#,maxOmegaThresh),
#CosmicCount=CosmicCount,
#maxOmegaCount= QFunclog(CosmicCount*maxOmega,maxOmegaCountThresh)
)
NoNewCtDNA <- setdiff(Master_Protein_Filtered_LR$BV.,NewCtDNA_ALLRounds$BV)
NoNewCtDNA_ALLRounds <- bind_rows(sapply(unique(NewCtDNA_ALLRounds$iteration),
FUN = function(x)
data.frame(BV=NoNewCtDNA,
iteration=x,
fold=as.numeric(balanced.cv.fold(NoNewCtDNA,
length(unique(NewCtDNA_ALLRounds$fold)))),
maxOmega=0),
simplify = F) )
NewCtDNA_ALLRounds_Modified <- rbind(NewCtDNA_ALLRounds,NoNewCtDNA_ALLRounds)
### RF Analysis#########
#### with CtDNA
OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData <-cv.CancervsNormal.QuantileReplace.Mutatation.knownCV(
CancerStatus =
1*(Master_Protein_Filtered_RF$Sample.Type !="Normal"),
Proteins =
OnlyProteins_RF %>%
dplyr::select(
CA199,
CEA,
CA125,
#AFP,
Prolactin,
HGF,
OPN ,
TIMP1,
#Follistatin,
#GCSF,
#HE4,
#CA153,
#IL6,
#Midkine,
Myeloperoxidase#,
#CYFRA211,
#Galectin3#,
#Thrombospondin2
) %>%
as.matrix() ,
ProteinsHigherCancer = T,
NewCtDNA_ALLRounds_Modified %>%
mutate(maxOmega=QFuncjumpramp(maxOmega,0)
#,
#maxOmegaCount=QFunclog(maxOmegaCount,Inf)
),
FeaturesToQuantile =
c("CA199",
"CEA",
"CA125",
"Prolactin",
"HGF",
"OPN" ,
"TIMP1",
#"IL6",
#"Midkine",
"Myeloperoxidase"),
QuantileFunction = function(x,Threshold)
ifelse(x>Threshold,x,0),
QuantileF = 0.95,
ForcePositiveCoeff = F,
NoLassoPenaly = T )
### Removing Protein Results
mainPath <- "Results/"
dir.create(mainPath,recursive = T)
sink(paste(mainPath,"ResultSummary.txt",sep = ""))
print("---------RF Data Results with 8 Proteins---------")
printRemoveDropResults(ThisResult = OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData)
ProbsRFData <- ProbsEachRound(OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData,
csvFile = paste(mainPath,"RFDatawithCtdna8proteins.csv",sep = "") )
LudaObject <- Generate4LudaObject(ThisResult = OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData,
rdaFile = paste(mainPath,"ForLuda8proteins.rda",sep = "") )
#Print scores
### Save Logit Scores
dir.create(paste(mainPath,"OnlyThisProtein/",sep = ""),recursive = T)
writeCSVMultiple(rownames(OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData$AllProbPredictionsMinusFeatureNIter) %>%
sapply(FUN = function(x) OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData$AllProbPredictionsMinusFeatureNIter[[x,"LR"]]%>%
mutate(Threshold=OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData$AllThresh99NIter[[x,"LR"]]),simplify = F),
file = paste(mainPath,"OnlyThisProtein/",sep = ""))
### Save Logit Scores
dir.create(paste(mainPath,"RemoveThisPrtoein/",sep = ""),recursive = T)
writeCSVMultiple(rownames(OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData$AllProbPredictionsDropFeatureNIter) %>%
sapply(FUN = function(x) OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData$AllProbPredictionsDropFeatureNIter[[x,"LR"]]%>%
mutate(Threshold=OnlyTenProteinsNoCosmicRemovedNewTemplatesRFData$AllThresh99NIter[[x,"LR"]]),simplify = F),
file = paste(mainPath,"RemoveThisPrtoein/",sep = ""))
### without CtdNA
OnlyTenProteinsNoCosmicRemovedNewTemplatesRFDatawithoutCtDNA <-cv.CancervsNormal.QuantileReplace.Mutatation.knownCV(
CancerStatus =
1*(Master_Protein_Filtered_RF$Sample.Type !="Normal"),
Proteins =
OnlyProteins_RF %>%
dplyr::select(
CA199,
CEA,
CA125,
#AFP,
Prolactin,
HGF,
OPN ,
TIMP1,
#Follistatin,
#GCSF,
#HE4,
#CA153,
#IL6,
#Midkine,
Myeloperoxidase#,
#CYFRA211,
#Galectin3#,
#Thrombospondin2
) %>%
as.matrix() ,
ProteinsHigherCancer = T,
NewCtDNA_ALLRounds_Modified %>%
mutate(maxOmega=QFuncjumpramp(maxOmega,Inf)
#,
#maxOmegaCount=QFunclog(maxOmegaCount,Inf)
),
FeaturesToQuantile =
c("CA199",
"CEA",
"CA125",
"Prolactin",
"HGF",
"OPN" ,
"TIMP1",
#"IL6",
#"Midkine",
"Myeloperoxidase"),
QuantileFunction = function(x,Threshold)
ifelse(x>Threshold,x,0),
QuantileF = 0.95,
ForcePositiveCoeff = F,
NoLassoPenaly = T )
printRemoveDropResults(ThisResult = OnlyTenProteinsNoCosmicRemovedNewTemplatesRFDatawithoutCtDNA)
ProbsRFData <- ProbsEachRound(OnlyTenProteinsNoCosmicRemovedNewTemplatesRFDatawithoutCtDNA,
csvFile = paste(mainPath,"RFDatawithCtdna8proteinswithoutCtdna.csv",sep = "") )
### Removing Each Protein at a time
EightProteins <- c("CA199",
"CEA",
"CA125",
"Prolactin",
"HGF",
"OPN" ,
"TIMP1",
"Myeloperoxidase")
SensOnceEightProteins <- setNames(rep(0,length(EightProteins)),EightProteins)
for(i in seq_along(EightProteins)){
EightProteinsRemoved <- setdiff(EightProteins,EightProteins[i])
OnlyTenProteinsNoCosmicRemovedNewTemplatesRFDataRemoveOneProtein <-
cv.CancervsNormal.QuantileReplace.Mutatation.knownCV(
CancerStatus =
1*(Master_Protein_Filtered_RF$Sample.Type !="Normal"),
Proteins =
OnlyProteins_RF[, EightProteinsRemoved] %>%
as.matrix() ,
ProteinsHigherCancer = T,
NewCtDNA_ALLRounds_Modified %>%
mutate(maxOmega=QFuncjumpramp(maxOmega,0)
#,
#maxOmegaCount=QFunclog(maxOmegaCount,Inf)
),
FeaturesToQuantile =
EightProteinsRemoved,
QuantileFunction = function(x,Threshold)
ifelse(x>Threshold,x,0),
QuantileF = 0.95,
ForcePositiveCoeff = F,
NoLassoPenaly = T )
ProbsRFData <- ProbsEachRound(OnlyTenProteinsNoCosmicRemovedNewTemplatesRFDataRemoveOneProtein,
csvFile = paste(mainPath,"RFDatawithCtdna8proteinswithout",EightProteins[i], ".csv",sep = "") )
SensOnceEightProteins[i] <- OnlyTenProteinsNoCosmicRemovedNewTemplatesRFDataRemoveOneProtein$AllSensitivity %>%
unlist() %>%
mean()
}
cat("--- Remove Protein------\n")
print(t(t(SensOnceEightProteins)))
sink()
Datasets
Models
