Using O’Connell’s ELISA data set introduced on the previous page, we prepare a data.frame for analysis.
In each of the tutorials, we will set up the R environment (or workspace) needed to perform analyses (and control output formatting). If tutorials are run one after the other some commands may be redundant, but reloading packages and preferences will not create any problems.
# Load the needed packages (in case not already done)
# recent versions of Rstudio load knitr as needed
# require(knitr)
# Plot formatting
source("AMgraph.R")## Loading required package: ggplot2
## Loading required package: ggthemes
## Loading required package: gridExtra
## Loading required package: grid
## Loading required package: plotrix# Data set tools
require(reshape2)## Loading required package: reshape2require(plyr)## Loading required package: plyrrequire(dplyr) ## Loading required package: dplyr
##
## Attaching package: 'dplyr'
##
## The following objects are masked from 'package:plyr':
##
## arrange, count, desc, failwith, id, mutate, rename, summarise,
## summarize
##
## The following object is masked from 'package:stats':
##
## filter
##
## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionrequire(grDevices)
# When working interactively (i.e. not calling this file from another
# function e.g. 'Knit' button or render() in outside script), set working
# directory to the location of this file. This way you will be working from
# the a common perspective---outside functions and paths will work either way.
# getwd()
# setwd("./source/R-new")
# Set options: output width, when scientific notation is used (see options help),
# do not include statistical significance star indicators
options(width=90, scipen = 4, show.signif.stars = FALSE)Data may be entered manually or imported from many different file formats (native Rdata, comma-separated, tab-delimited, etc).
Data sets can be stored as lists or rectangular tables (data.frames in R). Data.frames holding data that has some sort of repeating structure (groups, replicates) can be ‘long’ or ‘wide’. A long data.frame will have all the response data in one long column the calibrator concentration, group and replicate labels in other columns.
The following code shows how the O’Connell data may be entered manually in vector, or list, format then converted to a long data.frame. Later the long data.frame is converted to a wide one, which can be helpful for onscreen viewing or printing, or sharing data between R and other curve-fitting software such as GraphPad.
# O'Connell's ELISA data
# Enter the 12 calibrator concentrations
# Repeat sequence of known concentrations 3 times to correspond to 3 replicates
conc <- rep(c(0, 0, 3, 8, 23, 69, 206, 617, 1852, 5556, 16667, 50000), 3)
# Optical Density measurements
od <- c(
# Replicate 1
0.100, 0.120, 0.120, 0.120, 0.130, 0.153, 0.195, 0.280, 0.433, 0.593,
0.823, 0.933,
# Replicate 2
0.115, 0.110, 0.123, 0.118, 0.133, 0.160, 0.208, 0.305, 0.455, 0.668,
0.850, 1.078,
# Replicate 3
0.118, 0.115, 0.108, 0.118, 0.158, 0.150, 0.218, 0.323, 0.490, 0.760,
0.973, 0.825)
# Assign calibrator and replicate id just for easy identification
rep <- rep(c(1, 2, 3), each = 12)
calib <- rep(c("B1", "B2", paste("S0", seq(1:9), sep = ""), "S10"), 3)
# Combine vectors into a data.frame, 'ocon'.
# We will call this data.frame repeatedly in the steps below
ocon <- data.frame(calib, conc, rep, od)
# Print 1st 6 lines to console to check structure
head(ocon)## calib conc rep od
## 1 B1 0 1 0.100
## 2 B2 0 1 0.120
## 3 S01 3 1 0.120
## 4 S02 8 1 0.120
## 5 S03 23 1 0.130
## 6 S04 69 1 0.153# Or a more formally described structure:
str(ocon)## 'data.frame': 36 obs. of 4 variables:
## $ calib: Factor w/ 12 levels "B1","B2","S01",..: 1 2 3 4 5 6 7 8 9 10 ...
## $ conc : num 0 0 3 8 23 ...
## $ rep : num 1 1 1 1 1 1 1 1 1 1 ...
## $ od : num 0.1 0.12 0.12 0.12 0.13 0.153 0.195 0.28 0.433 0.593 ...# Save data as a .csv
# File path: "../" = go up one directory
# write.csv(ocon, file = "../data/ocon.csv", row.names = FALSE, quote = FALSE)Let’s summarise the data by concentration to get a ‘feel’ for our data set and check for errors:
# Mean response for concentration = 3
# Notice how the full data.frame$variable is needed each time a variable is named
# The argument na.rm=TRUE tells R to drop any missing observations
mean(ocon$od[ocon$conc == 3], na.rm = TRUE)## [1] 0.117sd(ocon$od[ocon$conc == 3], na.rm = TRUE)## [1] 0.007937254# Mean by calibrator concentration in one shot
ddply(ocon, .(conc), summarise,
Mean = mean(od, na.rm = TRUE),
SD = sd(od, na.rm = TRUE))## conc Mean SD
## 1 0 0.1130000 0.007211103
## 2 3 0.1170000 0.007937254
## 3 8 0.1186667 0.001154701
## 4 23 0.1403333 0.015373137
## 5 69 0.1543333 0.005131601
## 6 206 0.2070000 0.011532563
## 7 617 0.3026667 0.021594752
## 8 1852 0.4593333 0.028746014
## 9 5556 0.6736667 0.083644087
## 10 16667 0.8820000 0.079956238
## 11 50000 0.9453333 0.126950121Let’s examine minima and maxima for each concentration to check for extreme values:
oconMinMax <- ddply(ocon, .(conc), summarise,
n = length(na.omit(od)),
Min = min(od, na.rm = TRUE),
Max = max(od, na.rm = TRUE))
oconMinMax## conc n Min Max
## 1 0 6 0.100 0.120
## 2 3 3 0.108 0.123
## 3 8 3 0.118 0.120
## 4 23 3 0.130 0.158
## 5 69 3 0.150 0.160
## 6 206 3 0.195 0.218
## 7 617 3 0.280 0.323
## 8 1852 3 0.433 0.490
## 9 5556 3 0.593 0.760
## 10 16667 3 0.823 0.973
## 11 50000 3 0.825 1.078No glaring errors. We’ll examine the structure in more detail in the next 2 tutorials.
A wide, rather than long, format is nice for viewing and comparing calibrator replicates. Also, conversion between long and wide formats would be useful if one had to share data between R and other software such as GraphPad or a spreadsheet.
Prepare data set for GraphPad-type software:
# Reshape (aka transpose) long data set into wide format
oconWide <- dcast(ocon, calib + conc ~ rep , value.var = "od")
# Fix the variable names
names(oconWide)[3:5] <- paste("rep", names(oconWide[3:5]), sep = "")
# In this format, several runs could fit on one page.
oconWide## calib conc rep1 rep2 rep3
## 1 B1 0 0.100 0.115 0.118
## 2 B2 0 0.120 0.110 0.115
## 3 S01 3 0.120 0.123 0.108
## 4 S02 8 0.120 0.118 0.118
## 5 S03 23 0.130 0.133 0.158
## 6 S04 69 0.153 0.160 0.150
## 7 S05 206 0.195 0.208 0.218
## 8 S06 617 0.280 0.305 0.323
## 9 S07 1852 0.433 0.455 0.490
## 10 S08 5556 0.593 0.668 0.760
## 11 S09 16667 0.823 0.850 0.973
## 12 S10 50000 0.933 1.078 0.825# Limit the data to non-zeros and export it for later use in GraphPad
# write.csv(oconWide[3:12, 2:5], file.path(data.path, "ocon_GraphPad.csv"), quote = FALSE,
# row.names = FALSE, na = " ")Convert a wide data.frame to long format:
oconLong <- melt(oconWide, id = c("calib", "conc"))
head(oconLong, 15)## calib conc variable value
## 1 B1 0 rep1 0.100
## 2 B2 0 rep1 0.120
## 3 S01 3 rep1 0.120
## 4 S02 8 rep1 0.120
## 5 S03 23 rep1 0.130
## 6 S04 69 rep1 0.153
## 7 S05 206 rep1 0.195
## 8 S06 617 rep1 0.280
## 9 S07 1852 rep1 0.433
## 10 S08 5556 rep1 0.593
## 11 S09 16667 rep1 0.823
## 12 S10 50000 rep1 0.933
## 13 B1 0 rep2 0.115
## 14 B2 0 rep2 0.110
## 15 S01 3 rep2 0.123# Rename 'variable' to 'rep' and 'value' column to 'od'
names(oconLong)[c(2, 3)] <- c('rep', 'od')
str(oconLong)## 'data.frame': 36 obs. of 4 variables:
## $ calib: Factor w/ 12 levels "B1","B2","S01",..: 1 2 3 4 5 6 7 8 9 10 ...
## $ rep : num 0 0 3 8 23 ...
## $ od : Factor w/ 3 levels "rep1","rep2",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ value: num 0.1 0.12 0.12 0.12 0.13 0.153 0.195 0.28 0.433 0.593 ...Quick-R is a great resource for more details about entering data and R data types.
We have what we need to proceed to characterising the variance of the O’Connell data. For more practice with data entry and wrangling, see data preparation for R’s ELISA data set.