6 R

I could probably write 30 pages about my opinion on different R topics but I’ll spare you the pain of going over this here. All I want to do at this point, is to point out some of the excellent texts & tutorials that really push my own coding in R.

Then I need to mention 1/4 packages that anyone writing in R these days should be aware of. If you know them and choose not to use them - well that’s your issue. But if you have not looked at them so far - do it now! (Otherwise you’ll have serious issues when trying to follow any of my R code…)

The last issue I want to highlight is the use of standalone R scripts that are meant for used from the command line/ within a pipe. For me this is my attempt to create reproducible analysis that removes as much subjectivity as possible from the last part of the from raw data to figures journey.

6.1 Read

Anything written by Hadley Wickham will improve your R. He has written three great books (which are all publicly available for free)

If just start to get going with ggplot2:

Also, this platform of ggplot extensions is definitively worth a look.

6.2 Know

You should know the tidyverse. It not one, but actually a collection of packages which make working in R just so much more pleasant, streamlined, readable, fun…

The biggest impact so far on my own coding came from four packages that are part of the tidyverse. I list them below together with those functions from these packages which really made a difference.

magrittr: the pipe operator (%>%) is a game changer for R
ggplot2: simply the best way of data visualization I have used so far
dplyr: does for tables what ggplot2 does for plots (filter(), select(), mutate(), group_by(), summarise(),left_join(), tutorial)
purrr: this package changed the way I thought about code. Since I saw the beauty of map()/map2()/pmap() I pack most of my code into neat small functions. This saves soooooooo much typing/sources of error/confusion.

Of course there is many more great packages out there on CRAN, github or bioconductor. But the last package that I’m going to mention here is devtools. With this package you can easily install R packages from all different kinds of sources.

6.3 Run

Rstudio is a great environment to develop R code. I spend days using it and the idea of running plain R seems almost archaic to me. So during most of the time I will use R code throughout a project.

But when it comes to the final scripts that are going to make it to the publication, I want to drop as much interactivity as possible for the sake of reproducibility. I already keep this in mind during the process of writing my scripts. This means that every script should run from start to end without interactive intervention. No manual loading of data, interaction with other scripts basically only using source() (outsourcing definition of needed custom functions might actually be a good idea).

So, after polishing my R scripts to their final version they should be able to be run from the command line:

Rscript --vanilla script.R
# or
Rscript --vanilla script.R input_file.tsv input_value external_script.R

For this I provide the scripts that are meant for execution with a small header & config section (not the ones which I source()). Here is an example for the script that takes input from the command line.

#!/usr/bin/env Rscript
# run from terminal:
# Rscript --vanilla script.R input_file.tsv input_value external_script.R
# ===============================================================
# This script provides an example of an executable R script.
# It takes thee inputs and parses them into R variables
# ---------------------------------------------------------------
# The produced output contains:
#   - nothing, it's just a demo
# ===============================================================
# args contails everything thats typed into the command line -
# Our input starts with the 7th element.
# Careful: if you need more input fields, you also need to adjust the subsetting

args = commandArgs(trailingOnly = FALSE)
args = args[7:9]
print(args)

# Quick copy & paste section for devellopment phase
# args <- c('input_file.tsv', 'input_value', 'external_script.R')
# ------------------------------------
# Config:
# Set seed if your script uses any element of randomnes (sample(), rnorm(),...)

set.seed(42)                         
input_file <- as.character(args[1])
input_value <- as.numeric(args[2])
config_script <- as.character(args[3])

# ------------------------------------
# Here, the content of the original script starts 

library(tidyverse)
source(config_script)

data <- read_tsv(input_file)

# The code continues....