Using a vector as a variable in a formula - r

Question

I've got a function which acts on several columns, but I'd like to adapt it to use a different value of the main variable, mode, for each column. I have put a simplified example below.
My data is a cross tabulation of frequency i.e. in column A01 there are 6485 counts of 13 CAGs, 35 counts of 14 CAGs etc. The modal value for column 1 is therefore 13.
I need to calculate:
1) Skewness using (mean - mode)/sd
2) Proportion of each column where CAG is > than the mode
The code below works for that. However, I now need to compare each sample to the mode of a control sample and I'm a bit stuck with the code. The sample against which each needs to be compared is defined in the table controls. Could I please ask for help adapting my code so that skewmode and prop are calculated using the appropriate control mode for each column? I hope that makes sense!
#Data set
data <- data.frame(CAG = c(13, 14, 15, 17),
A01 = c(6485,35,132, 12),
A02 = c(0,42,56, 4))
#Mode
mode <- data[sapply(data[2:ncol(data)], which.max), ]$CAG
#Summary statistics
sumstats <- sapply(data[, 2:ncol(data)], function(x) {
data_e <- rep(data$CAG, x)
library(psych)
data.frame(
describe(data_e)
)
})
sumstats <- as.data.frame(t(sumstats))
sumstats[] <- lapply(sumstats, function(x) {
as.numeric(x)
})
# Results table
results <- data.frame(mode, sumstats)
# Skewness - I'd like to replace 'results$mode' here
# with the relevant mode from the controls table
skewmode <- (results$mean - results$mode) / results$sd
# Proportion > mode I'd like to replace 'mod' here
# with the relevant mode from the controls table
prop <- lapply(data[, 2:ncol(data)], function(x) {
mod <- data$CAG[which.max(x)]
B <- sum(x[data$CAG >= mod])
A <- sum(x[data$CAG <= mod])
B/(A+B)
})
prop <- as.data.frame(prop)
prop <- t(prop)
results <- data.frame(mode, sumstats, skewmode, prop)
# Controls
ctrls <- data.frame(samples = c('A01', 'A02', 'A03', 'A04'),
ctrl = c('A01','A01', 'A03', 'A03'))

Answer 1

Consider Map (the wrapper to mapply) which passes both sample mode and control mode iteratively into a defined function, prop_skew_calc(), to calculate skewmode and prop. At the end, outputs a list of dataframes for final row bind.
NOTE: below demonstrates with base R's summary() since I do not have the psyche package. However, I leave a comment in code on how to integrate psych::describe() which docs indicate returns a dataframe of summary stats useful for psychometrics:
Data (adding A03 and A04)
#Data set
data <- data.frame(CAG = c(13, 14, 15, 17),
A01 = c(6485,35,132, 12),
A02 = c(0,42,56, 4),
A03 = c(33,5014,2221, 18),
A04 = c(106,89,436, 11))
#Controls
ctrls <- data.frame(samples = c('A01', 'A02', 'A03', 'A04'),
ctrl = c('A01','A01', 'A03', 'A03'))
Function (removes any l/sapply looping since scalar values will be passed iteratively by Map)
library(psych)
prop_skew_calc <- function(x, y) {
#Mode
samplemode <- data$CAG[which.max(data[[x]])]
cntrlmode <- data$CAG[which.max(data[[y]])]
#Summary statistics
sumstats <- summary(rep(data$CAG, data[[x]])) # R base's summary()
sumstats <- as.data.frame(t(unclass(sumstats)))
#sumstats <- describe(rep(data$CAG, data[[x]])) # pysche's describe()
#sumstats <- as.data.frame(t(sumstats))
# Results table
results <- data.frame(cntrlmode, sumstats)
# Skewness
skewmode <- (results$Mean - results$cntrlmode) / results$Min
# Proportion
B <- sum(data[data$CAG >= cntrlmode, x])
A <- sum(data[data$CAG <= cntrlmode, x])
prop <- B/(A+B)
results <- data.frame(samplemode, cntrlmode, sumstats, skewmode, prop=prop)
}
Map (calling above function, passing columns of ctrl dataframe)
dfList <- Map(prop_skew_calc, ctrls$samples, ctrls$ctrl)
finaldf <- do.call(rbind, dfList)
finaldf
# samplemode cntrlmode Min. X1st.Qu. Median Mean X3rd.Qu. Max. skewmode prop
# 1 17 17 13 14 15 14.90 17 17 -0.1615385 0.223684211
# 2 13 17 13 13 13 13.05 13 17 -0.3038462 0.001797484
# 3 15 13 14 14 15 14.67 15 17 0.1192857 1.000000000
# 4 14 13 13 14 14 14.31 15 17 0.1007692 0.995491187