16
votes

My data is grouped by the IDs in V6 and ordered by position (V1:V3):

dt
      V1      V2      V3 V4 V5                 V6
 1: chr1 3054233 3054733  .  + ENSMUSG00000090025
 2: chr1 3102016 3102125  .  + ENSMUSG00000064842
 3: chr1 3205901 3207317  .  - ENSMUSG00000051951
 4: chr1 3206523 3207317  .  - ENSMUSG00000051951
 5: chr1 3213439 3215632  .  - ENSMUSG00000051951
 6: chr1 3213609 3216344  .  - ENSMUSG00000051951
 7: chr1 3214482 3216968  .  - ENSMUSG00000051951
 8: chr1 3421702 3421901  .  - ENSMUSG00000051951
 9: chr1 3466587 3466687  .  + ENSMUSG00000089699
10: chr1 3513405 3513553  .  + ENSMUSG00000089699

What I would like to do is to add and extra column with an index by position, that is, per group in V6 the first element would be "1", the second "2", and so on. I can achieve that using ddply and a custom function:

rankExons <- function(x){
   if(unique(x$V5) == "+"){ 
         x$index <- seq_len(nrow(x))}
   else{
         x$index <- rev(seq_len(nrow(x)))}
   x
}

indexed <- ddply(dt, .(V6), rankExons)
indexed
     V1      V2      V3 V4 V5                 V6 index
1  chr1 3205901 3207317  .  - ENSMUSG00000051951     6
2  chr1 3206523 3207317  .  - ENSMUSG00000051951     5
3  chr1 3213439 3215632  .  - ENSMUSG00000051951     4
4  chr1 3213609 3216344  .  - ENSMUSG00000051951     3
5  chr1 3214482 3216968  .  - ENSMUSG00000051951     2
6  chr1 3421702 3421901  .  - ENSMUSG00000051951     1
7  chr1 3102016 3102125  .  + ENSMUSG00000064842     1
8  chr1 3466587 3466687  .  + ENSMUSG00000089699     1
9  chr1 3513405 3513553  .  + ENSMUSG00000089699     2
10 chr1 3054233 3054733  .  + ENSMUSG00000090025     1

Unfortunately, it is extremely slow on the full dataset (~620k rows) and when using parallel it crashes and burns:

library(doMC)
registerDoMC(cores=6)
indexed <- ddply(dt, .(V6), rankExons, .parallel=TRUE)
Error: serialization is too large to store in a raw vector
Error: serialization is too large to store in a raw vector
Error: serialization is too large to store in a raw vector
Error: serialization is too large to store in a raw vector
Error: serialization is too large to store in a raw vector
Error: serialization is too large to store in a raw vector
Warning message:
In mclapply(argsList, FUN, mc.preschedule = preschedule, mc.set.seed = set.seed,  :
  all scheduled cores encountered errors in user code

So , I went for data.table but couldn't get it working. Here is what I tried:

setkey(dt, "V6")

dt[,index:=rankExons(dt), by=V6]
dt[,rankExons(.sd), by=V6, .SDcols=c("V5, V6")]

And both failed. How can I recreate my ddply with data.table?

dput(dt)
structure(list(V1 = c("chr1", "chr1", "chr1", "chr1", "chr1", 
"chr1", "chr1", "chr1", "chr1", "chr1"), V2 = c(3054233L, 3102016L, 
3205901L, 3206523L, 3213439L, 3213609L, 3214482L, 3421702L, 3466587L, 
3513405L), V3 = c(3054733L, 3102125L, 3207317L, 3207317L, 3215632L, 
3216344L, 3216968L, 3421901L, 3466687L, 3513553L), V4 = c(".", 
".", ".", ".", ".", ".", ".", ".", ".", "."), V5 = c("+", "+", 
"-", "-", "-", "-", "-", "-", "+", "+"), V6 = c("ENSMUSG00000090025", 
"ENSMUSG00000064842", "ENSMUSG00000051951", "ENSMUSG00000051951", 
"ENSMUSG00000051951", "ENSMUSG00000051951", "ENSMUSG00000051951", 
"ENSMUSG00000051951", "ENSMUSG00000089699", "ENSMUSG00000089699"
)), .Names = c("V1", "V2", "V3", "V4", "V5", "V6"), class = c("data.table", 
"data.frame"), row.names = c(NA, -10L), .internal.selfref = <pointer: 0x1de6a88>)
2
"Ask good questions, get good answers" should be the motto of stackoverflow :)fridaymeetssunday

2 Answers

27
votes

As a fellow bioinformatician, I come across this operation quite frequently. And this is where I adore data.table's modify subset of rows by reference feature!

I'd do it like this:

dt[V5 == "+", index := 1:.N, by=V6]
dt[V5 == "-", index := .N:1, by=V6]

No functions required. This is a little more advantageous because it avoids having to check for == "+" or "-" once for every group! Instead, you can first subset all groups with + once and then group by V6 and modify just those rows in place!

Similarly you do it once again for "-". Hope that helps.

Note: .N is a special variable that contains the number of observations per group.

3
votes

First, I'll load your sample data into R (you can't currently use dput() with data.table):

df <- read.table(header = TRUE, stringsAsFactors = FALSE, text = "
V1      V2      V3 V4 V5                 V6
1  chr1 3205901 3207317  .  - ENSMUSG00000051951
2  chr1 3206523 3207317  .  - ENSMUSG00000051951
3  chr1 3213439 3215632  .  - ENSMUSG00000051951
4  chr1 3213609 3216344  .  - ENSMUSG00000051951
5  chr1 3214482 3216968  .  - ENSMUSG00000051951
6  chr1 3421702 3421901  .  - ENSMUSG00000051951
7  chr1 3102016 3102125  .  + ENSMUSG00000064842
8  chr1 3466587 3466687  .  + ENSMUSG00000089699
9  chr1 3513405 3513553  .  + ENSMUSG00000089699
10 chr1 3054233 3054733  .  + ENSMUSG00000090025")

You can almost elegantly solve your problem with dplyr:

library(dplyr)

df %>% 
  group_by(V6, V5) %>%
  mutate(index = row_number(V2))

(I've assume V2 is the variable you want to index by - I think it's better to be explicit rather than relying on the order row of the row)

But you want a different summary for different subsets, which isn't currently easy in dplyr. One approach would be to split and then re-combine:

rbind_list(
  df %>% filter(V5 == "+") %>% mutate(index = row_number(V2)),
  df %>% filter(V5 == "-") %>% mutate(index = row_number(desc(V2)))
)

But this is going to be relatively slow since you have to make two copies of the data.

Another approach would to be use an if inside the summary:

df %>% 
  group_by(V6, V5) %>%
  mutate(index = row_number(if (V5[1] == "+") V2 else desc(V2)))