3

I want to calculate the area under the curve for several features measured at multiple concentrations for a group of subjects. The MESS auc function (described here: Calculate the Area under a Curve in R) gives me the auc, but I can't figure out to apply it to every column (feature) for all subjects in my data file.

My data is basically organized like this:

rowname  id      conc    feature1    feature2     feature3   ...
s1       ccr01   5       18575       80337        100496
s2       ccr01   4       18161       65723        109037
s3       ccr01   3       18092       99807        105363
s4       ccr01   2       5196        71520        84113
s5       ccr01   1       3940        50236        77145
s6       ccr02   5       1878        21812        10306
s7       ccr02   4       3660        18437        13408
s8       ccr02   3       4439        28379        25899
s9       ccr02   2       2710        22960        28080
s10      ccr02   1       1970        23557        22409
 .
 .
 .

I want to return a matrix/df of feature AUCs (columns) ordered by unique subject IDs (rows):

rowname    feature1    feature2    feature3
ccr01      52338.61    300823.6    388368.2
ccr02      12914.41    91486.32    84316.82

Any suggestions would be greatly appreciated!

Community
  • 1
  • 1
user2880691
  • 31
  • 4

1 Answers1

1

Using the function from the linked post and plyr to get the function ddply, this might work (and the data is named dat)

library(zoo)
AUC <- function(x, fs) 
    sapply(fs, function(f) sum(diff(x$conc)*rollmean(x[,f],2)))

library(plyr)
ddply(dat, .(id), function(x) {
    x <- x[order(x$conc),]
    AUC(x, grep("feature", names(x), value=T))
})

#      id feature1 feature2 feature3
# 1 ccr01  52706.5 302336.5 387333.5
# 2 ccr02  12733.0  92460.5  83744.5

Here, fs are the columns containing feature string, so it just applies the AUC function to those columns, grouped by id.

A dplyr solution,

library(dplyr)
AUC <- function(x, fs)
    setNames(as.data.frame(
        lapply(fs, function(f) sum(diff(x$conc)*rollmean(x[,f], 2)))), 
             fs)

dat %>% 
  group_by(id) %>%
  arrange(conc) %>%
  do(AUC(., grep("feature", names(.), value=T)))
Rorschach
  • 31,301
  • 5
  • 78
  • 129
  • Thanks, LegalizeIt. The AUC function with sapply works fine if I redefine fs as 3:ncol(x). However, I keep getting the following error when I try to use ddply: – user2880691 Jun 18 '15 at 03:51
  • is your data a matrix? In that case, you could change `names(x)` to `colnames(x)` or do as you did and use indices. what error? `ddply` won't work on the matrix, you could cast it to a `data.frame`, with `as.data.frame` if you want to use the `plyr` solution. – Rorschach Jun 18 '15 at 03:55
  • Thanks, LegalizeIt! The AUC function with sapply works fine if I redefine fs as 3:ncol(x). However, I keep getting the following error when I try to use ddply: Error: 'names' attribute [10] must be the same length as the vector [5] In this toy example, I have 2 IDs, 5 conc, and a total of 10 rows. What does .(id) do? – user2880691 Jun 18 '15 at 03:58
  • Sorry about that. I was installing packages that supported plyr. My toy data is a df, but I do have row names instead of numbered rows. Will that make a difference? – user2880691 Jun 18 '15 at 04:00
  • no the rownames shouldn't matter. The `.(id)` is just the syntax of that function, it means "group by id". – Rorschach Jun 18 '15 at 04:02
  • OK. Not sure why this is not working then. I'll keep banging on it. Thanks. – user2880691 Jun 18 '15 at 04:05
  • 1
    if you dump the output from `dput(head(yourData, 10))` in your original post I can test with your actual data. – Rorschach Jun 18 '15 at 04:06
  • OK. I've got the plyr version working now, which takes care of the tmp size limits for the local df. Thank you so much for your help, Legalizelt! I've got a few more bugs to work out with the real input data sets, but I think I'm almost there. – user2880691 Jun 18 '15 at 13:48