1
$\begingroup$

I am now struggling to do random forest analysis, I will be thankful if you could help with code for random forest analysis.

I got samples from the root, soil, and leaf from two regions (bau & mau) and these samples belong to two seasons (Wet and Dry).

Now I am interested to do random forest analysis at genera or family level to identify the taxa which contribute the differences like in root samples based on region as well as season.

Here is my code, but I am getting the error.

    library(randomForest)
    library(knitr)
    
    
    
    
    
    #### RANDOM FOREST ANALYSIS #####
     #### Prepare data ####
     #Load OTU table
     OTU_table=t(read.table("asv.table.txt", row.names=1,sep="\t", header=T, blank.lines.skip=F, check.names=F))
     table(apply(OTU_table,1,sum)) #verify rarefaction
    
     #Load metadata
     Meta=read.table("metadata.txt", header=T, row.names=1, stringsAsFactors=F, na.strings="NA",check.names=FALSE)
     Meta$sampleid=rownames(Meta)
    
     #Load taxonomy
     taxo=read.table("taxonomy.txt", row.names=1, sep="\t", header=F ,stringsAsFactors=F,quote="")
     rownames(taxo)=paste("a.",row.names(taxo),sep="")
    
    
     #### Run models ####
     #1. Root only
     # 1.1. both region
     # 1.2. bau only
     # 1.3. mau only
     #2. Soil only
    # 2.1. both region
     # 2.2. bau only
     # 2.3. mau only
     #Params RF
     NTREE=1000 # Number of Trees
     NbVar=1000 # Number of variables tested at each split
    
    
     
    
     
     #### Root ONLY 1-3 ####

 # 1. BOTH Region

 #Subset of data
 RootSamples=as.character(Meta[Meta$Compartment=="Root","sampleid"])
 Root_OTU_table=OTU_table[RootSamples,]




 #Model with microbiome based on Season, region

whole_root_pred=data.frame(Season=Meta[RootSamples,"Season"],Region=Meta[RootSamples,"Region"],a=Root_OTU_table)
 
head(whole_root_pred)
     Season Region a.d2ec9f3b77975c0f457e4b7413b217ff
     a.3147790f0d5a78316fb9dd64f53b9473 a.97aecc1f35cc1f50db507ad71dd22367
     a.bfad6370d28182cc6304844e9bec7fb6 a.5fa2a987221a1d9ca416148570c18086 


    **RF_model_Root_all=randomForest(y=?,sampsize=c(143,143),strata=?,x=whole_Root_pred,importance = T,proximity = T,ntree =
    NTREE,mtry = NbVar)**

print(RF_model_Root_all)
     #plot summary using the 5% most important OTUs ERROR ON LAST LINE
     imp=data.frame(importance(RF_model_Root_all))
     imp$genus=as.character(taxo[rownames(imp),"Genus"])
     Best=imp[imp$MeanDecreaseAccuracy>quantile(x = imp$MeanDecreaseAccuracy,.95),]
     bymedian <- with(Best, reorder(genus, -MeanDecreaseAccuracy, median))
    
     pdf(width = 20,height = 10,file=paste(pathforplots,"Variable_Importance_Root_BothRegion_raref.pdf",sep=""))
     par(mar=c(15,5,1,1))
     boxplot(Best$MeanDecreaseAccuracy ~ bymedian, data = Best,
     xlab = "", ylab = "Variable Importance",
     main = paste("Root in Both Countries; Error Rate=",round(RF_model_Feces_all$err.rate[NTREE,"OOB"],3),sep=""), varwidth = TRUE,
     col = "lightgray",las=2)
    
    
     dev.off()

Many thanks

$\endgroup$
2
$\begingroup$

This question will be a little hard to answer without more information.

For example, we will need to see your dataset (whole_root_pred), to decide why Stunting_Root is NULL.

  1. You might need to initialize Stunting_Root as a variable. It is currently not clear if it is e.g. a column of your dataframe, or just uninitialized. Uninitialized variables are NULL, which would explain your problem. randomForest might not know to look for strata inside your dataframe, for example. Is it in your dataframe?
  2. Also, I might be missing something, but why are you passing ? as a response? I'm not an expert but I believe that is an illegal character in R (I'm pretty sure?).
| improve this answer | |
$\endgroup$
  • $\begingroup$ Thank you so much for helping. I have updated the code. Actually after subset root data, I am not sure what to put in RF modal command in Y=? , as well as in strata?. Till whole_root_pred command, I am not getting any error but not sure after that what to do? $\endgroup$ – bioinfonext Oct 12 at 15:29
  • $\begingroup$ @bioinfonext I would recommend looking at the randomForest docs (?randomForest) to decide what to put in these arguments. y is supposed to be the response variable, so either a vector of categories (if task is classification) or a vector of numerics (if task is regression). Again, it would be very helpful to see a sample of the data, both the response variable and the predictor variables, and how they are represented in the code you provide. $\endgroup$ – Maximilian Press Oct 12 at 20:42
  • $\begingroup$ Thanks how can I show sample data! Is it head or dput command? $\endgroup$ – bioinfonext Oct 13 at 10:23
  • $\begingroup$ head() should be fine. $\endgroup$ – Maximilian Press Oct 13 at 18:57
  • $\begingroup$ I have added head(whole_root_pred) output. $\endgroup$ – bioinfonext Oct 15 at 13:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.