split character columns and get names of field in string

问题

I need to split a column that contains information into several columns.
I'd use tstrsplit but the same kind of information is not in the same order among the rows and I need to extract the name of the new column within the variable. Important to know: there can be many pieces of information (fields to become new variables) and I don't know all of them, so I don't want a "field by field" solution.

Below is an example of what I have:

library(data.table)

myDT <- structure(list(chr = c("chr1", "chr2", "chr4"), pos = c(123L,
                  435L, 120L), info = c("type=3;end=4", "end=6", "end=5;pos=TRUE;type=2"
                  )), class = c("data.table", "data.frame"), row.names = c(NA,-3L))

#    chr pos                  info
#1: chr1 123          type=3;end=4
#2: chr2 435                 end=6
#3: chr4 120 end=5;pos=TRUE;type=2

And I'd like to get:

#    chr pos end  pos type
#1: chr1 123   4 <NA>    3
#2: chr2 435   6 <NA> <NA>
#3: chr4 120   5 TRUE    2

A most straightforward way to get that would be much appreciated! (Note: I'm not willing to go with a dplyr/tidyr way)

回答1:

Using regex and the stringi packages:

setDT(myDT) # After creating data.table from structure()

library(stringi)

fields <- unique(unlist(stri_extract_all(regex = "[a-z]+(?==)", myDT$info)))
patterns <- sprintf("(?<=%s=)[^;]+", fields)
myDT[, (fields) := lapply(patterns, function(x) stri_extract(regex = x, info))]
myDT[, !"info"]

    chr  pos type end
1: chr1 <NA>    3   4
2: chr2 <NA> <NA>   6
3: chr4 TRUE    2   5

Edit: To get the correct type it seems (?) type.convert() can be used:

myDT[, (fields) := lapply(patterns, function(x) type.convert(stri_extract(regex = x, info), as.is = TRUE))]

回答2:

I am guessing your data is coming from a VCF file, if so there is a dedicated tool for such problems - bcftools.

Let's create example VCF file for testing:

# subset some data from 1000genomes data
tabix -h ftp://ftp-trace.ncbi.nih.gov/1000genomes/ftp/release/20100804/ALL.2of4intersection.20100804.genotypes.vcf.gz 17:1471000-1472000 > myFile.vcf
# zip it and index:
bgzip -c myFile.vcf > myFile.vcf.gz
tabix -p vcf myFile.vcf.gz

Now we can use bcftools. Here as an example we are subsetting AF and DP from INFO column:

bcftools query -f '%CHROM %POS %INFO/AF %INFO/DP \n' myFile.vcf.gz 
17  1471199  1916 0.088
17  1471538  2445 0.016
17  1471611  2733 0.239
17  1471623  2815 0.003
17  1471946  1608 0.007
17  1471959  1612 0.014
17  1471975  1610 0.179

See the manual for more query options.

回答3:

We could split on ";" then reshape wide-to-long, then split again on "=", then reshape back to long-to-wide:

dcast(
  melt(dt[,  paste0("col", 1:3) := tstrsplit(info, split = ";") ],
       id.vars = c("chr", "pos", "info"))[, -c("info", "variable")][
         ,c("x1", "x2") := tstrsplit(value, split = "=")][
           ,value := NULL][ !is.na(x1), ],
  chr + pos ~ x1, value.var = "x2")

#     chr pos end  pos type
# 1: chr1 123   4 <NA>    3
# 2: chr2 435   6 <NA> <NA>
# 3: chr4 120   5 TRUE    2

An improved / more readible version:

dt[, paste0("col", 1:3) := tstrsplit(info, split = ";")
   ][, melt(.SD, id.vars = c("chr", "pos", "info"), na.rm = TRUE)
     ][, -c("info", "variable")
       ][, c("x1", "x2") := tstrsplit(value, split = "=")
         ][, dcast(.SD, chr + pos ~ x1, value.var = "x2")]

回答4:

For now, I managed to get what I want with the following code:

newDT <- reshape(splitstackshape::cSplit(myDT, "info", sep=";", "long")[, 
                  c(.SD, tstrsplit(info, "="))], 
                 idvar=c("chr", "pos"), direction="wide", timevar="V4", drop="info")
setnames(newDT, sub("V5\\.", "", names(newDT)))

newDT
#    chr pos type end  pos
#1: chr1 123    3   4 <NA>
#2: chr2 435 <NA>   6 <NA>
#3: chr4 120    2   5 TRUE

Two options to improve the lines above, thanks to @A5C1D2H2I1M1N2O1R2T1 (who gave them in comments) :

. with a double cSplit prior to dcast:

cSplit(cSplit(myDT, "info", ";", "long"), "info", "=")[, dcast(.SD, chr + pos ~ info_1, value.var = "info_2")]

. with cSplit/trstrplit and dcast instead of reshape:

cSplit(myDT, "info", ";", "long")[, c("t1", "t2") := tstrsplit(info, "=", fixed = TRUE)][, dcast(.SD, chr + pos ~ t1, value.var = "t2")]

回答5:

Here's how I'd do it :

library(data.table)

myDT <- structure(list(chr = c("chr1", "chr2", "chr4"), pos = c(123L,
                                                                435L, 120L), info = c("type=3;end=4", "end=6", "end=5;pos=TRUE;type=2"
                                                                )), class = c("data.table", "data.frame"), row.names = c(NA,-3L))

R_strings <- paste0("list(", chartr(";", ",", myDT$info),")")
lists <- lapply(parse(text=R_strings),eval)
myDT[,info:=NULL]
myDT <- cbind(myDT,rbindlist(lists, fill = TRUE))
myDT
#>     chr pos type end  pos
#> 1: chr1 123    3   4   NA
#> 2: chr2 435   NA   6   NA
#> 3: chr4 120    2   5 TRUE

^{Created on 2019-11-29 by the reprex package (v0.3.0)}

回答6:

You could use separate calls to sub for each desired extracted field, e.g. for type:

myDT$type <- sub("^.*\\btype=([^;]+)\\b.*$", "\\1", myDT$info)

来源：https://stackoverflow.com/questions/59103539/split-character-columns-and-get-names-of-field-in-string

标签

data.table

reshape