How to subset data with advance string matching

Question

I have the following data frame from which I would like to extract rows based on matching strings.

> GEMA_EO5
gene_symbol  fold_EO  p_value                           RefSeq_ID      BH_p_value
       KNG1 3.433049 8.56e-28              NM_000893,NM_001102416    1.234245e-24
      REXO4 3.245317 1.78e-27                           NM_020385    2.281367e-24
      VPS29 3.827665 2.22e-25                 NM_057180,NM_016226    2.560770e-22
    CYP51A1 3.363149 5.95e-25              NM_000786,NM_001146152    6.239386e-22
      TNPO2 4.707600 1.60e-23 NM_001136195,NM_001136196,NM_013433    1.538000e-20
      NSDHL 2.703922 6.74e-23              NM_001129765,NM_015922    5.980454e-20
     DPYSL2 5.097382 1.29e-22                           NM_001386    1.062868e-19

So I would like to extract e.g. two rows based on matching strings in $RefSeq_ID, that works fine with the following:

> list<-c("NM_001386", "NM_020385")
> GEMA_EO6<-subset(GEMA_EO5, GEMA_EO5$RefSeq_ID %in% list, drop = TRUE)

> GEMA_EO6

gene_symbol  fold_EO  p_value RefSeq_ID    BH_p_value
      REXO4 3.245317 1.78e-27 NM_020385  2.281367e-24
     DPYSL2 5.097382 1.29e-22 NM_001386  1.062868e-19

But some of the rows have several RefSeq_IDs separated with commas, so I am looking for a general way of telling if $RefSeq_ID contains a certain string pattern and then subset that row.

Answer 1

To do partial matching you'll need to use regular expressions (see ?grepl). Here's a solution to your particular problem:

##Notice that the first element appears in 
##a row containing commas
l = c( "NM_013433", "NM_001386", "NM_020385")

To test one sequence at a time, we just select a particular seq id:

R> subset(GEMA_EO5, grepl(l[1], GEMA_EO5$RefSeq_ID))
  gene_symbol fold_EO p_value                           RefSeq_ID BH_p_value
5       TNPO2   4.708 1.6e-23 NM_001136195,NM_001136196,NM_013433  1.538e-20

To test for multiple genes, we use the | operator:

R> paste(l, collapse="|")
[1] "NM_013433|NM_001386|NM_020385"
R> grepl(paste(l, collapse="|"),GEMA_EO5$RefSeq_ID)
[1] FALSE  TRUE FALSE FALSE  TRUE FALSE  TRUE

So

subset(GEMA_EO5, grepl(paste(l, collapse="|"),GEMA_EO5$RefSeq_ID))

should give you what you want.

Answer 2

A different approach is to recognize the duplicate entries in RefSeq_ID as an attempt to represent two data base tables in a single data frame. So if the original table is csv, then normalize the data into two tables

Anno <- cbind(key = seq_len(nrow(csv)), csv[,names(csv) != "RefSeq_ID"])
key0 <- strsplit(csv$RefSeq_ID, ",")
RefSeq <- data.frame(key = rep(seq_along(key0), sapply(key0, length)),
                     ID = unlist(key0))

and recognize that the query is a subset (select) on the RefSeq table, followed by a merge (join) with Anno

l <- c( "NM_013433", "NM_001386", "NM_020385")
merge(Anno, subset(RefSeq, ID %in% l))[, -1]

leading to

> merge(Anno, subset(RefSeq, ID %in% l))[, -1]
  gene_symbol  fold_EO  p_value   BH_p_value        ID
1       REXO4 3.245317 1.78e-27 2.281367e-24 NM_020385
2       TNPO2 4.707600 1.60e-23 1.538000e-20 NM_013433
3      DPYSL2 5.097382 1.29e-22 1.062868e-19 NM_001386

Perhaps the goal is to merge with a `Master' table, then

Master <- cbind(key = seq_len(nrow(csv)), csv)
merge(Master, subset(RefSeq, ID %in% l))[,-1]

or similar.