Finding frequencies of all possible pairs in R

Question

I'm working with a large dataset of drugs and reactions using R. For now, I have the data structured as a very tall data frame that lists the report ID number, the Drug name, and the reported reactions. As you can tell, there is a one-to-many relationship between both IDs vs. drugs and drugs vs. reactions.

Keeping in mind that this dataset is MUCH larger than what I can duplicate here, I'd like to know how to find what pairs of drugs lead to what reactions and in what frequency .

Most importantly, I am interested in how to approach a problem like this. Is the data structured correctly? What concepts or libraries should I read about?

Here's a link to some real data: https://www.dropbox.com/s/kzx4mpyytbo9zil/query_result.csv

   ID    DRUG                                REACTION
1  1827  ASPIRIN                           CHEST PAIN
2  1827  CLARINEX                          CHEST PAIN
3  1827  ASPIRIN                                COUGH
4  1827  CLARINEX                               COUGH
5  1827  ASPIRIN                HAEMOGLOBIN DECREASED
6  1827  CLARINEX               HAEMOGLOBIN DECREASED
7  1827  ASPIRIN           NEUTROPHIL COUNT INCREASED
8  1827  CLARINEX          NEUTROPHIL COUNT INCREASED
9  1827  ASPIRIN               PHARYNGOLARYNGEAL PAIN
10 1827  CLARINEX              PHARYNGOLARYNGEAL PAIN
...

In my teeny little brain, the end result looks something like this...

    Drug1       Drug2       Reaction            Frequency
1   tylenol     alcohol     hepatic failure     298
2   advil       aleve       bleeding            201 
3   aspirin     advil       renal failure       199
4   docusate    senna       diarrhea            146
5   senna       sudafed     palpitations        121
6   xanax       alcohol     sedation            111
7   clarinex    benadryl    dry mouth           96
...
569 ASPIRIN     CLARINEX    CHEST PAIN          2

Drug1 and Drug2 are the drug pairs with the highest frequency from the entire dataset. A "drug pair" is defined as any combination of two drugs with the same report ID. The example output above would be interpreted as, "row 1 had 298 unique report IDs for which hepatic failure was the reaction."

Answer 1

Ok, I try an answer - I hope I got the question correctly. The code is rather intended to give some ideas than to be elegant/final.
Please note: I intentionally used for loops instead of possible vectorisation / apply functions, to make it easier to understand (those who are familiar with apply functions will also undertand the for loop ;-)).
Please note 2: Since I don't have more than a tiny piece of data, I could not test the code for the whole dataset!
EDIT : columns based on example above - possibly different from csv data.

Key points are:

• unique , [ etc.
• utils::combn to get combinations
• sum(FALSE/TRUE values) to count

Hope that helps!

require(utils)

df <- read.table(header=TRUE, 
text="LINE ID DRUG REACTION
1 1827 ASPIRIN CHEST_PAIN
2 1827 CLARINEX CHEST_PAIN
3 1827 ASPIRIN COUGH
4 1827 CLARINEX COUGH
5 1827 ASPIRIN HAEMOGLOBIN_DECREASED
6 1827 CLARINEX HAEMOGLOBIN_DECREASED
7 1827 ASPIRIN NEUTROPHIL_COUNT_INCREASED
8 1827 CLARINEX NEUTROPHIL_COUNT_INCREASED
9 1827 ASPIRIN PHARYNGOLARYNGEAL_PAIN
10 1827 CLARINEX PHARYNGOLARYNGEAL_PAIN")

# temporary object to collect if a combination is present
Results <- data.frame(Drug1=NA, Drug2=NA, Reaction=NA, Reaction.occurs=NA)
n=1 # start first line in Results object

#  walk through each ID ... 
for (ID in unique(df$ID)) { 

  # ... and each possible pair of drugs within a (report) ID ...
  drug.pairs <- utils::combn(x=unique(df[df$ID == ID, "DRUG"]), m=2) # the columns 
  for (ii in 1:ncol(drug.pairs)) {

    # ... and each reaction ...
    for (reaction in unique(df$REACTION)) {
      Results[n, "Drug1"] <- drug.pairs[1,ii]
      Results[n, "Drug2"] <- drug.pairs[2,ii]
      Results[n, "Reaction"] <- reaction
      Results[n, "Reaction.occurs"] <- drug.pairs[1,ii] %in% df[df$REACTION == reaction & df$ID == ID, "DRUG"] &
        drug.pairs[2,ii] %in% df[df$REACTION == reaction & df$ID == ID, "DRUG"]
      n <- n+1
    }
  }
}

head(Results)

# then find the unique Drug1 - Drug2 -Reaction combinations, and count the TRUE values:
(Results[!duplicated(Results[,1:3]), ][,1:3])
(unique(Results[, 1:3]))

# Results2 contains only the unique combinations
Results2 <- Results[!duplicated(Results[,1:3]), ][,1:3]

# calculatethe frequencies
for (i in 1:nrow(Results2)) {
  Results2[i, "Frequency"] <- sum(Results[Results$Drug1 == Results2[i, "Drug1"] & 
                                            Results$Drug2 == Results2[i, "Drug2"] & 
                                            Results$Reaction == Results2[i, "Reaction"], ]$Reaction.occurs)
}

Results2
# --- end ----

gives:

    Drug1    Drug2                   Reaction Frequency
1 ASPIRIN CLARINEX                 CHEST_PAIN         1
2 ASPIRIN CLARINEX                      COUGH         1
3 ASPIRIN CLARINEX      HAEMOGLOBIN_DECREASED         1
4 ASPIRIN CLARINEX NEUTROPHIL_COUNT_INCREASED         1
5 ASPIRIN CLARINEX     PHARYNGOLARYNGEAL_PAIN         1