R: Quickest way to create dataframe with an alternative to IFELSE

Question

I have a similar question this the one on this thread: Using R, replace all values in a matrix <0.1 with 0?

But in my case I have hypothetically larger dataset and variable thresholds. I need to create a dataframe with each value retrieved from a condition using the values on the first columns of the same dataframe. These values are different for each line.

Here is an example of the dataframe:

SNP        A1  A2   MAF     
rs3094315  G   A   0.172  
rs7419119  G   T   0.240  
rs13302957 G   A   0.081  
rs6696609  T   C   0.393 

Here is a sample of my code:

seqIndividuals = seq(1:201)
for(i in seqIndividuals) {
  alFrequ[paste("IND",i,"a",sep="")] = ifelse(runif(length(alFrequ$SNP),0.00,1.00) < alFrequ$MAF, alFrequ$A1, alFrequ$A2)
  alFrequ[paste("IND",i,"b",sep="")] = ifelse(runif(length(alFrequ$SNP),0.00,1.00) < alFrequ$MAF, alFrequ$A1, alFrequ$A2)
}

I am creating two new columns for each individual "i" in "seqIndividuals" by retrieving either values from column "A1" if a random value if lower than column "MAF", or "A2" if higher. The code is working great, but as a dataset grows in rows and columns (individuals) the time also grows significantly.

Is there a way to avoid using IFELSE for this situation, as I understand it works as a loop? I tried generating a matrix of random values and then replacing them, but it takes the same time or even longer.

mtxAlFrequ = matrix(runif(length(alFrequ$SNP)*(201)),nrow=length(alFrequ$SNP),ncol=201)
mtxAlFrequ[mtxAlFrequ < alFrequ$MAF] = alFrequ$A1

Thanks!

Answer 1

One option is data.table

library(data.table)
nm1 <- paste0("IND", rep(letters[1:2], length(seqIndividuals)), 
                    rep(seqIndividuals, each = 2))
setDT(alFrequ)
for(j in seq_along(nm1)) {
      alFrequ[, nm1[j] := A2
             ][runif(.N, 0, 1) < MAF , nm1[j] := A1][]
}

Benchmarks

set.seed(24)
alFrequ <- data.frame(SNP= paste0('rs', sample(600000, 340000, replace=FALSE)),
                   A1 = sample(c("G", "T", "A", "C"), 340000, replace=TRUE),
                   A2 = sample(c("G", "T", "A", "C"), 340000, replace=TRUE),
                   MAF = runif(340000, 0, 1), stringsAsFactors=FALSE)
nm1 <- paste0("IND", rep(letters[1:2], length(seqIndividuals)), 
                          rep(seqIndividuals, each = 2))

system.time({
    setDT(alFrequ)
     for(j in seq_along(nm1)){
     alFrequ[, nm1[j] := A2][runif(.N, 0, 1) < MAF , nm1[j] := A1][]
   }
})
#   user  system elapsed 
#  10.72    1.05   11.76 

and using the OP's code on the original dataset

system.time({
 for(i in seqIndividuals) {
   alFrequ[paste("IND",i,"a",sep="")] = ifelse(runif(length(alFrequ$SNP),0.00,1.00) < 
          alFrequ$MAF, alFrequ$A1, alFrequ$A2)
   alFrequ[paste("IND",i,"b",sep="")] = ifelse(runif(length(alFrequ$SNP),0.00,1.00) < 
             alFrequ$MAF, alFrequ$A1, alFrequ$A2)
 }
})
#    user  system elapsed 
#   72.16    6.82   79.33