How to match 2 dataframe columns and extract column values and column names?

Question

I have a matrix called mymat. I have a vector called geno <- c("01","N1","11","1N","10"). I have another table called key.table. What I want to do is I want to match the key column in key.table with the key column in mymat and If the column values in any of the matching rows have the any of the geno elements, I want to extract that column name from mymat along with the matching geno element and paste it in the new column in matched.extract in key.table in the corresponding rows for each key and get the result.

  mymat <- structure(c("chr5:12111", "chr5:12111", "chr5:12113", "chr5:12114", 
"chr5:12118", "0N", "0N", "1N", "0N", "0N", "00", "00", "00", 
"11", "10", "00", "00", "1N", "0N", "00"), .Dim = c(5L, 4L), .Dimnames = list(
    c("34", "35", "36", "37", "38"), c("key", "AMLM12001KP", 
    "AMAS-11.3-Diagnostic", "AMLM12014N-R")))

key.table<- structure(c("chr5:12111", "chr5:12111", "chr5:12113", "chr5:12114", 
"chr5:12118", "chr5:12122", "chr5:12123", "chr5:12123", "chr5:12125", 
"chr5:12127", "chr5:12129", "9920068", "9920069", "9920070", 
"9920071", "9920072", "9920073", "9920074", "9920075", "9920076", 
"9920077", "9920078"), .Dim = c(11L, 2L), .Dimnames = list(c("34", 
"35", "36", "37", "38", "39", "40", "41", "42", "43", "44"), 
    c("key", "variantId")))

result

  key          variantId    matched.extract
34 "chr5:12111" "9920068"     NA
35 "chr5:12111" "9920069"     NA
36 "chr5:12113" "9920070"     AMLM12001KP (1N),AMLM12014N-R (1N)
37 "chr5:12114" "9920071"     AMAS-11.3-Diagnostic (11)
38 "chr5:12118" "9920072"     AMAS-11.3-Diagnostic (10)
39 "chr5:12122" "9920073"     NA
40 "chr5:12123" "9920074"     NA
41 "chr5:12123" "9920075"     NA
42 "chr5:12125" "9920076"     NA
43 "chr5:12127" "9920077"     NA
44 "chr5:12129" "9920078"     NA

Answer 1

Using data.table, I would approach it like this:

library(data.table)
# convert the 'key.table' matrix to a data.table
kt <- as.data.table(key.table, keep.rownames=TRUE)
# convert the 'mymat' matrix to a data.table and melt into long format
# filter on the needed geno-types
# paste the needed values together into the requested format
mm <- melt(as.data.table(mymat, keep.rownames=TRUE),
           id=c("rn","key"))[value %in% c("1N","11","10"), val := paste0(variable," (",value,")")
                             ][, .(val = paste(val[!is.na(val)], collapse = ",")), by = .(rn,key)
                               ][val=="", val:=NA]
# join the 'mm' and 'kt' data.tables
kt[mm, matched := val, on=c("rn","key")]

which gives:

> kt
    rn        key variantId                            matched
 1: 34 chr5:12111   9920068                                 NA
 2: 35 chr5:12111   9920069                                 NA
 3: 36 chr5:12113   9920070 AMLM12001KP (1N),AMLM12014N-R (1N)
 4: 37 chr5:12114   9920071          AMAS-11.3-Diagnostic (11)
 5: 38 chr5:12118   9920072          AMAS-11.3-Diagnostic (10)
 6: 39 chr5:12122   9920073                                 NA
 7: 40 chr5:12123   9920074                                 NA
 8: 41 chr5:12123   9920075                                 NA
 9: 42 chr5:12125   9920076                                 NA
10: 43 chr5:12127   9920077                                 NA
11: 44 chr5:12129   9920078                                 NA

---

Explanation:

• kt <- as.data.table(key.table, keep.rownames=TRUE) will convert the matrix key.table to a data.table (which is an enhanced data.frame) and stores the rownames in the rn column.
• mm <- melt(as.data.table(mymat, keep.rownames=TRUE), id=c("rn","key")) will convert the matrix mymat to a data.table, stores the rownames in the rn column and melts the data.table into long format.
• the part [value %in% c("1N","11","10"), val := paste0(variable," (",value,")")] will paste the variable-values (which were the columnnams in mymat) with the value-values for only in the cases where value is 1N, 11 or 10.
• the part [, .(val = paste(val[!is.na(val)], collapse = ",")), by = .(rn,key)] will paste the non-NA rows of val together by the rn & key variables.
• the part [val=="", val:=NA] will transform the empty rows for val into NA-values
• finally kt[mm, matched := val, on=c("rn","key")] updates the kt-data.table by reference with the val-values of the mm-data.table for the matching rn & key variables.

---

_{WARNING: When using data.table, it is better not to use key as a variable name as key is also a parameter in a data.table. See ?key for more info.}

Answer 2

I'm not that familiar with the dplyr functions. You can try the base R merge function:

mm <- merge(key.table,mymat,by="key",all.x=T)
mm

function to paste the column names with the tissue type:

get.geno <- function(x,y) ifelse(!x %in% c("00","0N") & !is.na(x), paste0(y," (",x,")"), NA)
a <- t(apply(mm[,3:5], 1, get.geno, colnames(mm)[3:5]))

final dataframe:

mm$result <- apply(a, 1, function(x) paste(x[!is.na(x)] ,collapse=","))
mm[, -3:-5]
          key   variantId                           result
1  chr5:12111   9920068                                   
2  chr5:12111   9920068                                   
3  chr5:12111   9920069                                   
4  chr5:12111   9920069                                   
5  chr5:12113   9920070 AMLM12001KP (1N),AMLM12014N-R (1N)
6  chr5:12114   9920071          AMAS-11.3-Diagnostic (11)
7  chr5:12118   9920072          AMAS-11.3-Diagnostic (10)
8  chr5:12122   9920073                                   
9  chr5:12123   9920074                                   
10 chr5:12123   9920075                                   
11 chr5:12125   9920076                                   
12 chr5:12127   9920077                                   
13 chr5:12129   9920078