[R] predicting test dataset response from training dataset with randomForest

Mojca ZELNIKAR M.Zelnikar at sms.ed.ac.uk
Tue Aug 7 11:21:19 CEST 2012



Hi

I am new to R so I apologize if this is trivial.

I am trying to predict the resistance or susceptibility of my  
sequences to a certain drug with a randomForest function from a file  
with amino acids on each of the positions in the protein. I ran the  
following:

> library(randomForest)
>
> path <- "C:\\..."
> path2 <- "..."
> name <- "..."
>
> actualFileName <- paste(path, path2, name, ".txt", sep="")
>
> # reading in the training dataset
> dat1 <- read.table(actualFileName, header=TRUE, sep="\t",  
> colClasses="character")
>
> head(dat1)
   X1 X2 X3 X4 X5 X6 X7 X8 X9 X10 X11 X12 X13 X14 X15 X16 X17 ... SR
1  M  K  V  K  L  L  V  L  L   C   T   F   T   A   T   Y   A ... suscep
2  M  K  V  K  L  L  V  L  L   C   T   F   A   A   T   Y   A ... suscep
3  M  K  V  K  L  L  V  L  L   C   T   F   T   A   T   Y   A ... resist
...


> # some of the important sites identified by fisher test
> dat1$X13 <- as.factor(dat1$X13)
> dat1$X52 <- as.factor(dat1$X52)
> dat1$X53 <- as.factor(dat1$X53)
> dat1$X64 <- as.factor(dat1$X64)
> dat1$X85 <- as.factor(dat1$X85)
> dat1$X99 <- as.factor(dat1$X99)
> dat1$X111 <- as.factor(dat1$X111)
> dat1$X142 <- as.factor(dat1$X142)
> dat1$X157 <- as.factor(dat1$X157)
> dat1$X158 <- as.factor(dat1$X158)
> dat1$X162 <- as.factor(dat1$X162)
> dat1$X169 <- as.factor(dat1$X169)
> dat1$X200 <- as.factor(dat1$X200)
> dat1$X202 <- as.factor(dat1$X202)
> dat1$X203 <- as.factor(dat1$X203)
> dat1$X205 <- as.factor(dat1$X205)
> dat1$X206 <- as.factor(dat1$X206)
> dat1$X209 <- as.factor(dat1$X209)
> dat1$X210 <- as.factor(dat1$X210)
> dat1$X225 <- as.factor(dat1$X225)
> dat1$X269 <- as.factor(dat1$X269)
> dat1$X283 <- as.factor(dat1$X283)
> dat1$X290 <- as.factor(dat1$X290)
> dat1$X432 <- as.factor(dat1$X432)
> dat1$X434 <- as.factor(dat1$X434)
> dat1$X455 <- as.factor(dat1$X455)
> dat1$X467 <- as.factor(dat1$X467)
> dat1$X512 <- as.factor(dat1$X512)
> dat1$SR <- as.factor(dat1$SR)
>
>
>
> dat1.rf <-randomForest(SR ~ X13+ X52+ X53+ X64+ X85+ X99+ X111+  
> X142+ X157+ X158+ X162+ X169+ X200+
+ X202+ X203+ X205+ X206+ X209+ X210+ X225+ X269+ X283+ X290+ X432+  
X434+ X455+ X467+ X512, data=dat1, importance=TRUE,
+ proximity=TRUE, varUsed=TRUE, ntree=5000, varImpPlot=TRUE)
>
>
> print(dat1.rf)

> varImpPlot(dat1.rf)
>
> varUsed(dat1.rf, by.tree=FALSE, count=TRUE)

>
> MDSplot(dat1.rf, dat1$SR, palette=rep(1, 2),
+   pch=as.numeric(dat1$SR))
>
>
> path3 <- "C:\\Users..."
> path4 <- "..."
> name2 <- "..."
>
> # reading in the test dataset
> actualFileName2 <- paste(path3, path4, name2, ".txt", sep="")
>
> dat2 <- read.table(actualFileName2, header=TRUE, sep="\t",  
> colClasses="character")
>

>
> dat2$X13 <- as.factor(dat2$X13)
> dat2$X52 <- as.factor(dat2$X52)
> dat2$X53 <- as.factor(dat2$X53)
> dat2$X64 <- as.factor(dat2$X64)
> dat2$X85 <- as.factor(dat2$X85)
> dat2$X99 <- as.factor(dat2$X99)
> dat2$X111 <- as.factor(dat2$X111)
> dat2$X142 <- as.factor(dat2$X142)
> dat2$X157 <- as.factor(dat2$X157)
> dat2$X158 <- as.factor(dat2$X158)
> dat2$X162 <- as.factor(dat2$X162)
> dat2$X169 <- as.factor(dat2$X169)
> dat2$X200 <- as.factor(dat2$X200)
> dat2$X202 <- as.factor(dat2$X202)
> dat2$X203 <- as.factor(dat2$X203)
> dat2$X205 <- as.factor(dat2$X205)
> dat2$X206 <- as.factor(dat2$X206)
> dat2$X209 <- as.factor(dat2$X209)
> dat2$X210 <- as.factor(dat2$X210)
> dat2$X225 <- as.factor(dat2$X225)
> dat2$X269 <- as.factor(dat2$X269)
> dat2$X283 <- as.factor(dat2$X283)
> dat2$X290 <- as.factor(dat2$X290)
> dat2$X432 <- as.factor(dat2$X432)
> dat2$X434 <- as.factor(dat2$X434)
> dat2$X455 <- as.factor(dat2$X455)
> dat2$X467 <- as.factor(dat2$X467)
> dat2$X512 <- as.factor(dat2$X512)
> dat2$SR <- as.factor(dat2$SR)
>
>
>
> dat2.pred<-predict(dat1.rf, dat2, type="response", norm.votes=TRUE,  
> predict.all=FALSE, proximity=FALSE, nodes=FALSE)

Error in predict.randomForest(dat1.rf, dat2, type = "response",  
norm.votes = TRUE,  :
   New factor levels not present in the training data
>
The thing is that each of the amino acid positions in the training  
dataset is present also in the training dataset. So I don't know how  
to deal with the error.

Thank you very much.

Kind regards,

Mojca Zelnikar

-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.



More information about the R-help mailing list