changeset 4:830e8dc3acac draft default tip

Uploaded
author jfb
date Tue, 14 Jul 2020 19:51:12 -0400
parents 6eacb089b655
children
files C and D finder/CandD.xml C and D finder/commonality-working feb 21 20.R C and D finder/commonality-working jan 13 2020.R C and D finder/commonality-working jul 13 2020.R
diffstat 4 files changed, 258 insertions(+), 588 deletions(-) [+]
line wrap: on
line diff
--- a/C and D finder/CandD.xml	Mon Feb 24 17:54:14 2020 -0500
+++ b/C and D finder/CandD.xml	Tue Jul 14 19:51:12 2020 -0400
@@ -11,7 +11,7 @@
 		ln -s '$input5' input5.csv &&
 		ln -s '$input6' input6.csv &&
 		
-		Rscript '$__tool_directory__/commonality-working feb 21 20.R'
+		Rscript '$__tool_directory__/commonality-working jul 13 2020.R'
     ]]></command>
     <inputs>
 		<param format="csv" name="input1" type="data" label="First Substrate Set"/>
--- a/C and D finder/commonality-working feb 21 20.R	Mon Feb 24 17:54:14 2020 -0500
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,302 +0,0 @@
-FirstSubstrateSet<- read.csv("input1.csv", stringsAsFactors=FALSE, header = FALSE)
-Firstsubbackfreq<- read.csv("input2.csv", header=FALSE, stringsAsFactors=FALSE)
-SubstrateHeader<-FirstSubstrateSet[1,]
-FirstSubstrateSet<- FirstSubstrateSet[2:nrow(FirstSubstrateSet),]
-if(nrow(Firstsubbackfreq[1,]>35)){
-  if(grepl(pattern = "Properties", x=Firstsubbackfreq[1,22])){
-    Firstsubbackfreq<-t(Firstsubbackfreq)
-  }
-}
-
-
-SecondSubstrateSet<- read.csv("input3.csv", stringsAsFactors=FALSE, header = FALSE)
-Secondsubbackfreq<- read.csv("input4.csv", header=FALSE, stringsAsFactors=FALSE)
-SecondSubstrateSet<- SecondSubstrateSet[2:nrow(SecondSubstrateSet),]
-if(nrow(Secondsubbackfreq[1,]>35)){
-  if(grepl(pattern = "Properties", x=Secondsubbackfreq[1,22])){
-    Secondsubbackfreq<-t(Secondsubbackfreq)
-  }
-}
-
-
-ThirdSubstrateSet<- read.csv("input5.csv", stringsAsFactors=FALSE, header = FALSE)
-Thirdsubbackfreq<- read.csv("input6.csv", header=FALSE, stringsAsFactors=FALSE)
-ThirdSubstrateSet<- ThirdSubstrateSet[2:nrow(ThirdSubstrateSet),]
-if(nrow(Thirdsubbackfreq[1,]>35)){
-  if(grepl(pattern = "Properties", x=Thirdsubbackfreq[1,22])){
-    Thirdsubbackfreq<-t(Thirdsubbackfreq)
-  }
-}
-
-
-
-
-#ff you want ONLY FULL MOTIFS, put "YES" here, please use all caps
-FullMotifsOnly_questionmark<-"NO"
-#If you want ONLY TRUNCATED MOTIFS, put "YES" here, please use all caps
-TruncatedMotifsOnly_questionmark<-"NO"
-#if you want to find the overlap, put a "YES" here (all caps), if you want to find the non-overlap, put "NO" (all caps)
-Are_You_Looking_For_Commonality<-"YES"
-
-
-#then put the names of your output files here
-Shared_motifs_table<-"sharedmotifs.csv"
-Shared_subbackfreq_table<-"sharedSBF.csv"
-
-# Shared_motifs_table<-"Shared motifs 7-27-17.csv"
-# Shared_subbackfreq_table<-"SubstrateBackgrounFrequency-for-shared-motifs 4 7-27-17.csv"
-
-First_unshared_motifs_table<-"R1 substrates.csv"
-First_unshared_subbackfreq<-"R1 SBF.csv"
-
-Second_unshared_motifs_table<-"R2 subs.csv"
-Second_unshared_subbackfreq<-"R2 SBf.csv"
-
-Third_unshared_motifs_table<-"R3 subs.csv"
-Third_unshared_subbackfreq<-"R3 SBF.csv"
-
-#final note, this code is going to be unworkable if you want to make a Venn diagram of more than 3 circles.  I think I'll poke around
-#other languages to see if any of them can do it.
-####################################################################################################################################
-
-
-
-
-
-FirstxY<-rep("xY",times=nrow(FirstSubstrateSet))
-FirstSubstrateSet[,11]<-FirstxY
-
-SecondxY<-rep("xY",times=nrow(SecondSubstrateSet))
-SecondSubstrateSet[,11]<-SecondxY
-
-ThirdxY<-rep("xY",times=nrow(ThirdSubstrateSet))
-ThirdSubstrateSet[,11]<-ThirdxY
-
-
-
-
-
-
-
-
-
-
-
-####################################################################################################################################
-####################################################################################################################################
-# better version of this code written in C: what happens when two kinases share a motif, but they found that motif in two 
-# separate proteins thus two separate accession numbers?
-# It should actually output the shared motif and BOTH accession numbers.  Right now it does not, it only maps out the second
-# accession number.  So that needs to be fixed BUT you need to keep the commonality between a motif and its accession number
-####################################################################################################################################
-####################################################################################################################################
-####################################################################################################################################
-####################################################################################################################################
-
-#Create the motif sets, deciding wether or not you're looking for truncated or full here
-#full only
- 
-FTLwtmotifs=matrix(,nrow = nrow(FirstSubstrateSet),ncol=1)
-FTLwtAccessionNumbers=matrix(data = Firstsubbackfreq[1,],ncol=1)
-
-for (i in 1:nrow(FirstSubstrateSet)){
-  FTLwtletters<-FirstSubstrateSet[i,4:18]
-  FTLwtletters<-FTLwtletters[FTLwtletters !="XXXXX"]
-  FTLwtletters<-paste(FTLwtletters, sep="", collapse="")
-  leftspaces<-c()
-  rightspaces<-c()
-  
-  YYYmotif <- unlist(strsplit(FTLwtletters, split = ""))
-  YYYposition <- match(x = "x", table = YYYmotif)
-  #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
-  #just 3 letters to the left of x
-  
-  YYYLettersToTheLeft <- YYYposition - 1
-  #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
-  #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
-  YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
-  #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
-  #variable the user puts in is
-  
-  
-  if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
-    leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
-    rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
-    #add blank spaces if the motif has less than 4 letters to the left/right
-    motif<-c(leftspaces,YYYmotif,rightspaces)
-    #save that motif, which is the Y and +/- 4 amino acids, including truncation
-    motif<-motif[!motif %in% "x"]
-    motif<-paste(motif, sep="", collapse="")
-    FTLwtletters<-motif
-    FTLwtmotifs[i,1]<-FTLwtletters
-    # FTLwtAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-  }
-  
-  if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
-    motif<-YYYmotif
-    #add blank spaces if the motif has less than 4 letters to the left/right
-    motif<-c(leftspaces,YYYmotif,rightspaces)
-    #save that motif, which is the Y and +/- 4 amino acids, including truncation
-    motif<-motif[!motif %in% "x"]
-    motif<-paste(motif, sep="", collapse="")
-    FTLwtletters<-motif
-    FTLwtmotifs[i,1]<-FTLwtletters
-    # FTLwtAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-    
-    
-  }
-  
-}
-
-D835Ymotifs=matrix(,nrow = nrow(SecondSubstrateSet),ncol=1)
-D835YAccessionNumbers<-matrix(data = Secondsubbackfreq[1,],ncol = 1)
-
-for (i in 1:nrow(SecondSubstrateSet)){
-  D835letters<-SecondSubstrateSet[i,4:18]
-  D835letters<-D835letters[D835letters !="XXXXX"]
-  D835letters<-paste(D835letters, sep="", collapse="")
-  leftspaces<-c()
-  rightspaces<-c()
-  
-  YYYmotif <- unlist(strsplit(D835letters, split = ""))
-  YYYposition <- match(x = "x", table = YYYmotif)
-  #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
-  #just 3 letters to the left of x
-  
-  YYYLettersToTheLeft <- YYYposition - 1
-  #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
-  #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
-  YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
-  #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
-  #variable the user puts in is
-  if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
-    leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
-    rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
-    #add blank spaces if the motif has less than 4 letters to the left/right
-    motif<-c(leftspaces,YYYmotif,rightspaces)
-    #save that motif, which is the Y and +/- 4 amino acids, including truncation
-    motif<-motif[!motif %in% "x"]
-    motif<-paste(motif, sep="", collapse="")
-    D835letters<-motif
-    D835Ymotifs[i,1]<-D835letters
-    # D835YAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-  }
-  
-  if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
-    motif<-YYYmotif
-    #add blank spaces if the motif has less than 4 letters to the left/right
-    motif<-c(leftspaces,YYYmotif,rightspaces)
-    #save that motif, which is the Y and +/- 4 amino acids, including truncation
-    motif<-motif[!motif %in% "x"]
-    motif<-paste(motif, sep="", collapse="")
-    D835letters<-motif
-    D835Ymotifs[i,1]<-D835letters
-    # D835YAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-  }
-}
-
-
-ITDmotifs=matrix(,nrow = nrow(ThirdSubstrateSet),ncol=1)
-ITDAccessionNumbers<-matrix(data = Thirdsubbackfreq[1,],ncol = 1)
-
-for (i in 1:nrow(ThirdSubstrateSet)){
-  ITDletters<-ThirdSubstrateSet[i,4:18]
-  ITDletters<-ITDletters[ITDletters !="XXXXX"]
-  ITDletters<-paste(ITDletters, sep="", collapse="")
-  YYYmotif <- unlist(strsplit(ITDletters, split = ""))
-  leftspaces<-c()
-  rightspaces<-c()
-  YYYposition <- match(x = "x", table = YYYmotif)
-  #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
-  #just 3 letters to the left of x
-  
-  YYYLettersToTheLeft <- YYYposition - 1
-  #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
-  #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
-  YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
-  #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
-  #variable the user puts in is
-  if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
-    leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
-    rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
-    #add blank spaces if the motif has less than 4 letters to the left/right
-    motif<-c(leftspaces,YYYmotif,rightspaces)
-    #save that motif, which is the Y and +/- 4 amino acids, including truncation
-    motif<-motif[!motif %in% "x"]
-    motif<-paste(motif, sep="", collapse="")
-    ITDletters<-motif
-    ITDmotifs[i,1]<-ITDletters
-    # ITDAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-  }
-  
-  if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
-    motif<-YYYmotif
-    #add blank spaces if the motif has less than 4 letters to the left/right
-    motif<-c(leftspaces,YYYmotif,rightspaces)
-    #save that motif, which is the Y and +/- 4 amino acids, including truncation
-    motif<-motif[!motif %in% "x"]
-    motif<-paste(motif, sep="", collapse="")
-    ITDletters<-motif
-    ITDmotifs[i,1]<-ITDletters
-    # ITDAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-  }
-}
-  
-#############################################################################################################################
-#############################################################################################################################
-#############################################################################################################################
-#############################################################################################################################
-#############################################################################################################################
-
-#now look for either commonality or difference.  Actually could you look for both...
-
-if (Are_You_Looking_For_Commonality=="YES"){
-  
-  columnalheader<-c(as.character(Thirdsubbackfreq[1:36,1]))
-  columnalheader<-matrix(columnalheader,nrow = 1)
-
-  SubstrateOverlap1<-intersect(D835Ymotifs,ITDmotifs)
-  SubstrateOverlap1<-as.matrix(SubstrateOverlap1)
-  
-  
-  columnalheader<-c(rep(NA,36))
-  FinalMatrix<-matrix(data =columnalheader,ncol = 1)
-
-  SubstrateOverlapFINAL<-intersect(FTLwtmotifs,SubstrateOverlap1)
-  AccessionOverlap1<-intersect(D835YAccessionNumbers,ITDAccessionNumbers)
-  AccessionOverlapFinal<-intersect(AccessionOverlap1,FTLwtAccessionNumbers)
-  AccessionOverlapFinal<-unlist(AccessionOverlapFinal)
-  
-  for (x in 1:length(AccessionOverlapFinal)) {
-    for (y in 1:ncol(Firstsubbackfreq)) {
-      Acc<-AccessionOverlapFinal[x]
-      SBF<-Firstsubbackfreq[1,y]
-      if(Acc==SBF){
-        FinalMatrix<-cbind(FinalMatrix,Firstsubbackfreq[,y])
-      }
-    }
-  }
-  FinalMatrix<-FinalMatrix[,2:ncol(FinalMatrix)]
-  
-  if(grepl(pattern = "Properties", x=FinalMatrix[22,1])==FALSE){
-    Outputmatrix<-cbind(Firstsubbackfreq[,1],FinalMatrix)
-    write.table(x=Outputmatrix,file = Shared_subbackfreq_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")
-  } else {
-    write.table(x=FinalMatrix,file = Shared_subbackfreq_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")
-  }
-  
-  SubstrateMatrix<-SubstrateHeader
-  if(ncol(SubstrateMatrix)>18){
-    SubstrateMatrix<-SubstrateMatrix[,1:18]
-  }
-  
-  for (z in 1:length(SubstrateOverlapFINAL)) {
-    motif<-SubstrateOverlapFINAL[z]
-    newmotif<-unlist(strsplit(motif,split = ""))
-    
-    Addition<-""
-    outputmotif<-c(Addition,Addition,Addition,newmotif)
-    SubstrateMatrix<-rbind(SubstrateMatrix,outputmotif)
-  }
-  write.table(x=SubstrateMatrix,file = Shared_motifs_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")
-}
--- a/C and D finder/commonality-working jan 13 2020.R	Mon Feb 24 17:54:14 2020 -0500
+++ /dev/null	Thu Jan 01 00:00:00 1970 +0000
@@ -1,285 +0,0 @@
-FirstSubstrateSet<- read.csv("input1.csv", stringsAsFactors=FALSE, header = FALSE)
-Firstsubbackfreq<- read.csv("input2.csv", header=FALSE, stringsAsFactors=FALSE)
-SubstrateHeader<-FirstSubstrateSet[1,]
-FirstSubstrateSet<- FirstSubstrateSet[2:nrow(FirstSubstrateSet),]
-
-SecondSubstrateSet<- read.csv("input3.csv", stringsAsFactors=FALSE, header = FALSE)
-Secondsubbackfreq<- read.csv("input4.csv", header=FALSE, stringsAsFactors=FALSE)
-SecondSubstrateSet<- SecondSubstrateSet[2:nrow(SecondSubstrateSet),]
-
-
-ThirdSubstrateSet<- read.csv("input5.csv", stringsAsFactors=FALSE, header = FALSE)
-Thirdsubbackfreq<- read.csv("input6.csv", header=FALSE, stringsAsFactors=FALSE)
-ThirdSubstrateSet<- ThirdSubstrateSet[2:nrow(ThirdSubstrateSet),]
-
-
-#ff you want ONLY FULL MOTIFS, put "YES" here, please use all caps
-FullMotifsOnly_questionmark<-"NO"
-#If you want ONLY TRUNCATED MOTIFS, put "YES" here, please use all caps
-TruncatedMotifsOnly_questionmark<-"NO"
-#if you want to find the overlap, put a "YES" here (all caps), if you want to find the non-overlap, put "NO" (all caps)
-Are_You_Looking_For_Commonality<-"YES"
-
-
-#then put the names of your output files here
-Shared_motifs_table<-"sharedmotifs.csv"
-Shared_subbackfreq_table<-"sharedSBF.csv"
-
-# Shared_motifs_table<-"Shared motifs 7-27-17.csv"
-# Shared_subbackfreq_table<-"SubstrateBackgrounFrequency-for-shared-motifs 4 7-27-17.csv"
-
-First_unshared_motifs_table<-"R1 substrates.csv"
-First_unshared_subbackfreq<-"R1 SBF.csv"
-
-Second_unshared_motifs_table<-"R2 subs.csv"
-Second_unshared_subbackfreq<-"R2 SBf.csv"
-
-Third_unshared_motifs_table<-"R3 subs.csv"
-Third_unshared_subbackfreq<-"R3 SBF.csv"
-
-#final note, this code is going to be unworkable if you want to make a Venn diagram of more than 3 circles.  I think I'll poke around
-#other languages to see if any of them can do it.
-####################################################################################################################################
-
-
-
-
-
-FirstxY<-rep("xY",times=nrow(FirstSubstrateSet))
-FirstSubstrateSet[,11]<-FirstxY
-
-SecondxY<-rep("xY",times=nrow(SecondSubstrateSet))
-SecondSubstrateSet[,11]<-SecondxY
-
-ThirdxY<-rep("xY",times=nrow(ThirdSubstrateSet))
-ThirdSubstrateSet[,11]<-ThirdxY
-
-
-
-
-
-
-
-
-
-
-
-####################################################################################################################################
-####################################################################################################################################
-# better version of this code written in C: what happens when two kinases share a motif, but they found that motif in two 
-# separate proteins thus two separate accession numbers?
-# It should actually output the shared motif and BOTH accession numbers.  Right now it does not, it only maps out the second
-# accession number.  So that needs to be fixed BUT you need to keep the commonality between a motif and its accession number
-####################################################################################################################################
-####################################################################################################################################
-####################################################################################################################################
-####################################################################################################################################
-
-#Create the motif sets, deciding wether or not you're looking for truncated or full here
-#full only
-if (Are_You_Looking_For_Commonality=="YES"){
-
-  
-  ###############################################
-  #ALL motifs, full and truncated
-  
-    FTLwtmotifs=matrix(,nrow = nrow(FirstSubstrateSet),ncol=1)
-    FTLwtAccessionNumbers=matrix(data = Firstsubbackfreq[1,],ncol=1)
-    
-    for (i in 1:nrow(FirstSubstrateSet)){
-      FTLwtletters<-FirstSubstrateSet[i,4:18]
-      FTLwtletters<-FTLwtletters[FTLwtletters !="XXXXX"]
-      FTLwtletters<-paste(FTLwtletters, sep="", collapse="")
-      leftspaces<-c()
-      rightspaces<-c()
-      
-      YYYmotif <- unlist(strsplit(FTLwtletters, split = ""))
-      YYYposition <- match(x = "x", table = YYYmotif)
-      #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
-      #just 3 letters to the left of x
-      
-      YYYLettersToTheLeft <- YYYposition - 1
-      #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
-      #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
-      YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
-      #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
-      #variable the user puts in is
-      
-      
-      if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
-        leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
-        rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
-        #add blank spaces if the motif has less than 4 letters to the left/right
-        motif<-c(leftspaces,YYYmotif,rightspaces)
-        #save that motif, which is the Y and +/- 4 amino acids, including truncation
-        motif<-motif[!motif %in% "x"]
-        motif<-paste(motif, sep="", collapse="")
-        FTLwtletters<-motif
-        FTLwtmotifs[i,1]<-FTLwtletters
-        # FTLwtAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-      }
-      
-      if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
-        motif<-YYYmotif
-        #add blank spaces if the motif has less than 4 letters to the left/right
-        motif<-c(leftspaces,YYYmotif,rightspaces)
-        #save that motif, which is the Y and +/- 4 amino acids, including truncation
-        motif<-motif[!motif %in% "x"]
-        motif<-paste(motif, sep="", collapse="")
-        FTLwtletters<-motif
-        FTLwtmotifs[i,1]<-FTLwtletters
-        # FTLwtAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-        
-        
-      }
-      
-    }
-    
-    D835Ymotifs=matrix(,nrow = nrow(SecondSubstrateSet),ncol=1)
-    D835YAccessionNumbers<-matrix(data = Secondsubbackfreq[1,],ncol = 1)
-    
-    for (i in 1:nrow(SecondSubstrateSet)){
-      D835letters<-SecondSubstrateSet[i,4:18]
-      D835letters<-D835letters[D835letters !="XXXXX"]
-      D835letters<-paste(D835letters, sep="", collapse="")
-      leftspaces<-c()
-      rightspaces<-c()
-      
-      YYYmotif <- unlist(strsplit(D835letters, split = ""))
-      YYYposition <- match(x = "x", table = YYYmotif)
-      #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
-      #just 3 letters to the left of x
-      
-      YYYLettersToTheLeft <- YYYposition - 1
-      #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
-      #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
-      YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
-      #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
-      #variable the user puts in is
-      if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
-        leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
-        rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
-        #add blank spaces if the motif has less than 4 letters to the left/right
-        motif<-c(leftspaces,YYYmotif,rightspaces)
-        #save that motif, which is the Y and +/- 4 amino acids, including truncation
-        motif<-motif[!motif %in% "x"]
-        motif<-paste(motif, sep="", collapse="")
-        D835letters<-motif
-        D835Ymotifs[i,1]<-D835letters
-        D835YAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-      }
-      
-      if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
-        motif<-YYYmotif
-        #add blank spaces if the motif has less than 4 letters to the left/right
-        motif<-c(leftspaces,YYYmotif,rightspaces)
-        #save that motif, which is the Y and +/- 4 amino acids, including truncation
-        motif<-motif[!motif %in% "x"]
-        motif<-paste(motif, sep="", collapse="")
-        D835letters<-motif
-        D835Ymotifs[i,1]<-D835letters
-        D835YAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-      }
-    }
-    
-    
-    ITDmotifs=matrix(,nrow = nrow(ThirdSubstrateSet),ncol=1)
-    ITDAccessionNumbers<-matrix(data = Thirdsubbackfreq[1,],ncol = 1)
-    
-    for (i in 1:nrow(ThirdSubstrateSet)){
-      ITDletters<-ThirdSubstrateSet[i,4:18]
-      ITDletters<-ITDletters[ITDletters !="XXXXX"]
-      ITDletters<-paste(ITDletters, sep="", collapse="")
-      YYYmotif <- unlist(strsplit(ITDletters, split = ""))
-      leftspaces<-c()
-      rightspaces<-c()
-      YYYposition <- match(x = "x", table = YYYmotif)
-      #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
-      #just 3 letters to the left of x
-      
-      YYYLettersToTheLeft <- YYYposition - 1
-      #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
-      #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
-      YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
-      #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
-      #variable the user puts in is
-      if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
-        leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
-        rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
-        #add blank spaces if the motif has less than 4 letters to the left/right
-        motif<-c(leftspaces,YYYmotif,rightspaces)
-        #save that motif, which is the Y and +/- 4 amino acids, including truncation
-        motif<-motif[!motif %in% "x"]
-        motif<-paste(motif, sep="", collapse="")
-        ITDletters<-motif
-        ITDmotifs[i,1]<-ITDletters
-        ITDAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-      }
-      
-      if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
-        motif<-YYYmotif
-        #add blank spaces if the motif has less than 4 letters to the left/right
-        motif<-c(leftspaces,YYYmotif,rightspaces)
-        #save that motif, which is the Y and +/- 4 amino acids, including truncation
-        motif<-motif[!motif %in% "x"]
-        motif<-paste(motif, sep="", collapse="")
-        ITDletters<-motif
-        ITDmotifs[i,1]<-ITDletters
-        ITDAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
-      }
-    }
-    
-  #############################################################################################################################
-  #############################################################################################################################
-  #############################################################################################################################
-  #############################################################################################################################
-  #############################################################################################################################
-  
-  #now look for either commonality or difference.  Actually could you look for both...
-  
-  if (Are_You_Looking_For_Commonality=="YES"){
-    
-    columnalheader<-c(as.character(Thirdsubbackfreq[1:36,1]))
-    columnalheader<-matrix(columnalheader,nrow = 1)
-
-    SubstrateOverlap1<-intersect(D835Ymotifs,ITDmotifs)
-    SubstrateOverlap1<-as.matrix(SubstrateOverlap1)
-    
-    
-    columnalheader<-c(rep(NA,36))
-    FinalMatrix<-matrix(data =columnalheader,ncol = 1)
-
-    SubstrateOverlapFINAL<-intersect(FTLwtmotifs,SubstrateOverlap1)
-    AccessionOverlap1<-intersect(D835YAccessionNumbers,ITDAccessionNumbers)
-    AccessionOverlapFinal<-intersect(AccessionOverlap1,FTLwtAccessionNumbers)
-    AccessionOverlapFinal<-unlist(AccessionOverlapFinal)
-    
-    for (x in 1:length(AccessionOverlapFinal)) {
-      for (y in 1:ncol(Firstsubbackfreq)) {
-        Acc<-AccessionOverlapFinal[x]
-        SBF<-Firstsubbackfreq[1,y]
-        if(Acc==SBF){
-          FinalMatrix<-cbind(FinalMatrix,Firstsubbackfreq[,y])
-        }
-      }
-    }
-    FinalMatrix<-FinalMatrix[,2:ncol(FinalMatrix)]
-    Outputmatrix<-cbind(Firstsubbackfreq[,1],FinalMatrix)
-    write.table(x=Outputmatrix,file = Shared_subbackfreq_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")
-    
-    SubstrateMatrix<-SubstrateHeader
-    if(ncol(SubstrateMatrix)>18){
-      SubstrateMatrix<-SubstrateMatrix[,1:18]
-    }
-    
-    for (z in 1:length(SubstrateOverlapFINAL)) {
-      motif<-SubstrateOverlapFINAL[z]
-      newmotif<-unlist(strsplit(motif,split = ""))
-      
-      Addition<-""
-      outputmotif<-c(Addition,Addition,Addition,newmotif)
-      SubstrateMatrix<-rbind(SubstrateMatrix,outputmotif)
-    }
-    write.table(x=SubstrateMatrix,file = Shared_motifs_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")
-  }
-}
\ No newline at end of file
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/C and D finder/commonality-working jul 13 2020.R	Tue Jul 14 19:51:12 2020 -0400
@@ -0,0 +1,257 @@
+FirstSubstrateSet<- read.csv("input1.csv", stringsAsFactors=FALSE, header = FALSE)
+Firstsubbackfreq<- read.csv("input2.csv", header=FALSE, stringsAsFactors=FALSE)
+SubstrateHeader<-FirstSubstrateSet[1,]
+FirstSubstrateSet<- FirstSubstrateSet[2:nrow(FirstSubstrateSet),]
+if(nrow(Firstsubbackfreq[1,]>35)){
+  if(grepl(pattern = "Properties", x=Firstsubbackfreq[1,22])){
+    Firstsubbackfreq<-t(Firstsubbackfreq)
+  }
+}
+
+
+SecondSubstrateSet<- read.csv("input3.csv", stringsAsFactors=FALSE, header = FALSE)
+Secondsubbackfreq<- read.csv("input4.csv", header=FALSE, stringsAsFactors=FALSE)
+SecondSubstrateSet<- SecondSubstrateSet[2:nrow(SecondSubstrateSet),]
+if(nrow(Secondsubbackfreq[1,]>35)){
+  if(grepl(pattern = "Properties", x=Secondsubbackfreq[1,22])){
+    Secondsubbackfreq<-t(Secondsubbackfreq)
+  }
+}
+
+
+ThirdSubstrateSet<- read.csv("input5.csv", stringsAsFactors=FALSE, header = FALSE)
+Thirdsubbackfreq<- read.csv("input6.csv", header=FALSE, stringsAsFactors=FALSE)
+ThirdSubstrateSet<- ThirdSubstrateSet[2:nrow(ThirdSubstrateSet),]
+if(nrow(Thirdsubbackfreq[1,]>35)){
+  if(grepl(pattern = "Properties", x=Thirdsubbackfreq[1,22])){
+    Thirdsubbackfreq<-t(Thirdsubbackfreq)
+  }
+}
+
+#the above three sections bring in the input files and ensure they are properly aligned.  The if statements align them if they are misaligned
+
+First_unshared_motifs_table<-"R1 substrates.csv"
+First_unshared_subbackfreq<-"R1 SBF.csv"
+
+Second_unshared_motifs_table<-"R2 subs.csv"
+Second_unshared_subbackfreq<-"R2 SBf.csv"
+
+Third_unshared_motifs_table<-"R3 subs.csv"
+Third_unshared_subbackfreq<-"R3 SBF.csv"
+
+#the above 4 sections create the names of the output files that this tool can create
+
+
+
+
+FirstxY<-rep("xY",times=nrow(FirstSubstrateSet))
+FirstSubstrateSet[,11]<-FirstxY
+
+SecondxY<-rep("xY",times=nrow(SecondSubstrateSet))
+SecondSubstrateSet[,11]<-SecondxY
+
+ThirdxY<-rep("xY",times=nrow(ThirdSubstrateSet))
+ThirdSubstrateSet[,11]<-ThirdxY
+
+#for each input file, mark the phospho-Tyrosine (which is always housed in column 11) 
+#with an xY to denote that is it a phospho and not regular tyrosine
+
+
+#currently the substrates are a dataframe with many values, I want to collapse each substrate into a single variable, 
+#but if I simply use the paste() function then I lose any information about whether the substrate was truncated on the C or N terminal.  
+#so the for loops below then are constructed so as to retain that information on how any substrate is truncated.
+
+FTLwtmotifs=matrix(,nrow = nrow(FirstSubstrateSet),ncol=1)
+FTLwtAccessionNumbers=matrix(data = Firstsubbackfreq[1,],ncol=1)
+#create the vectors which will house the first set of substrates and the first set of accession numbers
+
+
+for (i in 1:nrow(FirstSubstrateSet)){
+  FTLwtletters<-FirstSubstrateSet[i,4:18]
+  FTLwtletters<-FTLwtletters[FTLwtletters !="XXXXX"]
+  FTLwtletters<-paste(FTLwtletters, sep="", collapse="")
+  leftspaces<-c()
+  rightspaces<-c()
+  
+  YYYmotif <- unlist(strsplit(FTLwtletters, split = ""))
+  YYYposition <- match(x = "x", table = YYYmotif)
+  #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
+  #just 3 letters to the left of x
+  
+  YYYLettersToTheLeft <- YYYposition - 1
+  YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
+  #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
+  #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
+  
+  
+  
+  if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
+    leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
+    rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
+    #add blank spaces if the motif has less than 4 letters to the left/right
+    motif<-c(leftspaces,YYYmotif,rightspaces)
+    #save that motif, which is the Y and +/- 4 amino acids, including truncation
+    motif<-motif[!motif %in% "x"]
+    motif<-paste(motif, sep="", collapse="")
+    FTLwtletters<-motif
+    FTLwtmotifs[i,1]<-FTLwtletters
+  }
+  
+  if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
+    motif<-YYYmotif
+    #add blank spaces if the motif has less than 4 letters to the left/right
+    motif<-c(leftspaces,YYYmotif,rightspaces)
+    #save that motif, which is the Y and +/- 4 amino acids, including truncation
+    motif<-motif[!motif %in% "x"]
+    motif<-paste(motif, sep="", collapse="")
+    FTLwtletters<-motif
+    FTLwtmotifs[i,1]<-FTLwtletters
+  }
+}
+
+D835Ymotifs=matrix(,nrow = nrow(SecondSubstrateSet),ncol=1)
+D835YAccessionNumbers<-matrix(data = Secondsubbackfreq[1,],ncol = 1)
+#vectors to house the second set of substrates and accession numbers
+
+for (i in 1:nrow(SecondSubstrateSet)){
+  D835letters<-SecondSubstrateSet[i,4:18]
+  D835letters<-D835letters[D835letters !="XXXXX"]
+  D835letters<-paste(D835letters, sep="", collapse="")
+  leftspaces<-c()
+  rightspaces<-c()
+  
+  YYYmotif <- unlist(strsplit(D835letters, split = ""))
+  YYYposition <- match(x = "x", table = YYYmotif)
+  #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
+  #just 3 letters to the left of x
+  
+  YYYLettersToTheLeft <- YYYposition - 1
+  #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
+  #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
+  YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
+  #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
+  #variable the user puts in is
+  if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
+    leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
+    rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
+    #add blank spaces if the motif has less than 4 letters to the left/right
+    motif<-c(leftspaces,YYYmotif,rightspaces)
+    #save that motif, which is the Y and +/- 4 amino acids, including truncation
+    motif<-motif[!motif %in% "x"]
+    motif<-paste(motif, sep="", collapse="")
+    D835letters<-motif
+    D835Ymotifs[i,1]<-D835letters
+  }
+  
+  if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
+    motif<-YYYmotif
+    #add blank spaces if the motif has less than 4 letters to the left/right
+    motif<-c(leftspaces,YYYmotif,rightspaces)
+    #save that motif, which is the Y and +/- 4 amino acids, including truncation
+    motif<-motif[!motif %in% "x"]
+    motif<-paste(motif, sep="", collapse="")
+    D835letters<-motif
+    D835Ymotifs[i,1]<-D835letters
+  }
+}
+
+
+ITDmotifs=matrix(,nrow = nrow(ThirdSubstrateSet),ncol=1)
+ITDAccessionNumbers<-matrix(data = Thirdsubbackfreq[1,],ncol = 1)
+#third set of substrates and accession numbers
+
+
+for (i in 1:nrow(ThirdSubstrateSet)){
+  ITDletters<-ThirdSubstrateSet[i,4:18]
+  ITDletters<-ITDletters[ITDletters !="XXXXX"]
+  ITDletters<-paste(ITDletters, sep="", collapse="")
+  YYYmotif <- unlist(strsplit(ITDletters, split = ""))
+  leftspaces<-c()
+  rightspaces<-c()
+  YYYposition <- match(x = "x", table = YYYmotif)
+  #position itself tells me how much is to the left of that X by what it's number is.  x at position 4 tells me that there are
+  #just 3 letters to the left of x
+  
+  YYYLettersToTheLeft <- YYYposition - 1
+  #how many letters to the right SHOULD just be length(motif)-position-1 if it's 5 long and x is at 3 then Y is at 4 and there is
+  #just 1 spot to the right of Y so LettersToTheRight<-1 because 5-3-1=1
+  YYYLettersToTheRight <- length(YYYmotif) - YYYposition - 1
+  #then sanity check, we're currently looking only at +/-4, but this spot allows for up to +/- 7 as well, just depends on what the
+  #variable the user puts in is
+  if (YYYLettersToTheLeft < 7 | YYYLettersToTheRight < 7) {
+    leftspaces<-rep(" ",times=(7-YYYLettersToTheLeft))
+    rightspaces<-rep(" ",times=7-(YYYLettersToTheRight))
+    #add blank spaces if the motif has less than 4 letters to the left/right
+    motif<-c(leftspaces,YYYmotif,rightspaces)
+    #save that motif, which is the Y and +/- 4 amino acids, including truncation
+    motif<-motif[!motif %in% "x"]
+    motif<-paste(motif, sep="", collapse="")
+    ITDletters<-motif
+    ITDmotifs[i,1]<-ITDletters
+    # ITDAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
+  }
+  
+  if(YYYLettersToTheLeft>6 && YYYLettersToTheRight>6){
+    motif<-YYYmotif
+    #add blank spaces if the motif has less than 4 letters to the left/right
+    motif<-c(leftspaces,YYYmotif,rightspaces)
+    #save that motif, which is the Y and +/- 4 amino acids, including truncation
+    motif<-motif[!motif %in% "x"]
+    motif<-paste(motif, sep="", collapse="")
+    ITDletters<-motif
+    ITDmotifs[i,1]<-ITDletters
+    # ITDAccessionNumbers[i,1]<-FirstSubstrateSet[i,3]
+  }
+}
+
+
+SubstrateOverlap1<-intersect(D835Ymotifs,ITDmotifs)
+SubstrateOverlap1<-as.matrix(SubstrateOverlap1)
+SubstrateOverlapFINAL<-intersect(FTLwtmotifs,SubstrateOverlap1)
+#this tool is essentially looking for the intersection of three sets, and to do so it performs 1 intersection at a time
+#this find the intersection of the substrate sets
+
+AccessionOverlap1<-intersect(D835YAccessionNumbers,ITDAccessionNumbers)
+AccessionOverlapFinal<-intersect(AccessionOverlap1,FTLwtAccessionNumbers)
+AccessionOverlapFinal<-unlist(AccessionOverlapFinal)
+#this tool is essentially looking for the intersection of three sets, and to do so it performs 1 intersection at a time
+#this find the intersection of the accession number sets
+
+
+for (x in 1:length(AccessionOverlapFinal)) {
+  for (y in 1:ncol(Firstsubbackfreq)) {
+    Acc<-AccessionOverlapFinal[x]
+    SBF<-Firstsubbackfreq[1,y]
+    if(Acc==SBF){
+      FinalMatrix<-cbind(FinalMatrix,Firstsubbackfreq[,y])
+    }
+  }
+}
+#for every accession number, go back to the original substrate background frequency file and find the protein statistics associated with that number
+#columnbind all those protein statistics together
+FinalMatrix<-FinalMatrix[,2:ncol(FinalMatrix)]
+
+
+#write all files, in the proper format so that the files coming out look like the files coming in
+
+if(grepl(pattern = "Properties", x=FinalMatrix[22,1])==FALSE){
+  Outputmatrix<-cbind(Firstsubbackfreq[,1],FinalMatrix)
+  write.table(x=Outputmatrix,file = Shared_subbackfreq_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")
+} else {
+  write.table(x=FinalMatrix,file = Shared_subbackfreq_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")
+}
+
+SubstrateMatrix<-SubstrateHeader
+if(ncol(SubstrateMatrix)>18){
+  SubstrateMatrix<-SubstrateMatrix[,1:18]
+}
+
+for (z in 1:length(SubstrateOverlapFINAL)) {
+  motif<-SubstrateOverlapFINAL[z]
+  newmotif<-unlist(strsplit(motif,split = ""))
+  
+  Addition<-""
+  outputmotif<-c(Addition,Addition,Addition,newmotif)
+  SubstrateMatrix<-rbind(SubstrateMatrix,outputmotif)
+}
+write.table(x=SubstrateMatrix,file = Shared_motifs_table,quote = FALSE,sep = ",",row.names = FALSE,col.names = FALSE,na="")