Loop to perform calculations across rows on specific columns matching a pattern (in data frame)?

问题

I have a dataframe with some boolean values (1/0) as follows (sorry I couldn't work out how to make this into a smart table)

       Flag1.Sam Flag2.Sam Flag3.Sam Flag1.Ted Flag2.Ted Flag3.Ted
probe1         0         1         0         1         0         0
probe2         0         0         0         0         0         0
probe3         1         0         0         0         0         0
probe4         0         0         0         0         0         0
probe5         1         1         0         1         0         0

I have 64 samples (Sam/Ted....etc) which are in a list called files i.e;

files <- c("Sam", "Ted", "Ann", ....)

And I would like to create a a column summing the flag values for each sample to create the following:

               Sam Ted 
probe1.flagsum   1   1
probe2.flagsum   0   0 
probe3.flagsum   1   0 
probe4.flagsum   0   0
probe5.flagsum   2   1

I am fairly new to R, trying to learn on a need to know basis but I have tried the following:

for(i in files) {
    FLAGS$i <- cbind(sapply(i, function(y) { 
        #greping columns to filter for one sample
        filter1 <- grep(names(filters), pattern=y)
        #print out the summed values for those columns  
        FLAGS$y <-rowSums(filters[,(filter1)])
    }
}

The above code does not work and I am bit lost as how to move forward.

Can anyone help me untangle this problem or point me in the right direction of the commands/tools to use.

Thank you.

回答1:

This is easily doable in base R reshape, though using the reshape or reshape2 packages might be more intuitive.

Here's a solution in base R:

# Here's your data in its current form
dat = read.table(header=TRUE, text="Flag1.Sam Flag2.Sam   Flag3.Sam   Flag1.Ted   Flag2.Ted   Flag3.Ted
probe1 0   1   0   1   0   0
probe2 0   0   0   0   0   0
probe3 1   0   0   0   0   0
probe4 0   0   0   0   0   0
probe5 1   1   0   1   0   0")
# Generate an ID row
dat$id = row.names(dat)
# Reshape wide to long
r.dat = reshape(dat, direction="long", 
                timevar="probe", 
                varying=1:6, sep=".")
# Calculate row sums
r.dat$sum = rowSums(r.dat[3:5])
# Reshape back to wide format, dropping what you're not interested in
reshape(r.dat, direction="wide", 
        idvar="id", timevar="probe", 
        drop=3:5)
##                id sum.Sam sum.Ted
## probe1.Sam probe1       1       1
## probe2.Sam probe2       0       0
## probe3.Sam probe3       1       0
## probe4.Sam probe4       0       0
## probe5.Sam probe5       2       1

More than one way to skin a cat

You can also whip up a function like this one:

myFun = function(data, varnames) {
  temp = vector("list", length(varnames))
  for (i in 1:length(varnames)) {
    temp[[i]] = colSums(t(dat[grep(varnames[i], names(data))]))
    names(temp)[[i]] = varnames[i]
  }
  data.frame(temp)
}

Then, making use of the vector that you have of names:

files = c("Sam", "Ted")
myFun(dat, files)
##        Sam Ted
## probe1   1   1
## probe2   0   0
## probe3   1   0
## probe4   0   0
## probe5   2   1

Enjoy!

回答2:

If filters is your input matrix and FLAGS your desired output matrix then I would (naïvely) do something like this:

FLAGS <- matrix(0,nrow=nrow(filters),ncol=length(files))
for(i in 1:length(files)){
    grep(files[i],colnames(filters)) -> index
    FLAGS[,i] <- rowSums(filters[,index])
    }
colnames(FLAGS) <- files

回答3:

assuming your matrix is called input

input <- matrix(rbinom(30, 1, 0.5), ncol = 6)
colnames(input) <- c("F1.S", "F2.S", "F3.S", "F1.T", "F2.T", "F3.T")
rownames(input) <- paste("probe", 1:5, sep = "")
input <- as.data.frame(input)

library(reshape)
input$probe <- rownames(input)
Molten <- melt(input, id.vars = "probe")
Molten$ID <- gsub("^.*\\.", "", levels(Molten$variable))[Molten$variable]
cast(probe ~ ID, data = Molten, fun = "sum")

update with the dat frame from mrdwab

dat = read.table(header=TRUE, text="Flag1.Sam Flag2.Sam   Flag3.Sam   Flag1.Ted   Flag2.Ted   Flag3.Ted
probe1 0   1   0   1   0   0
probe2 0   0   0   0   0   0
probe3 1   0   0   0   0   0
probe4 0   0   0   0   0   0
probe5 1   1   0   1   0   0")

library(reshape)
dat$probe <- rownames(dat)
Molten <- melt(dat, id.vars = "probe")
Molten$ID <- gsub("^.*\\.", "", levels(Molten$variable))[Molten$variable]
cast(probe ~ ID, data = Molten, fun = "sum")

来源：https://stackoverflow.com/questions/11118384/loop-to-perform-calculations-across-rows-on-specific-columns-matching-a-pattern

标签

for-loop

pattern-matching

aggregate

reshape