R: Level 2 (Intermediate)

1. Picking out exactly the data you want

In R Basics you pulled a column with $. Two more ways to select are everywhere in real analysis: by a name and by a condition.

2. Matrices and lists: the other building blocks

A matrix is a grid of one type (think an expression matrix). A list can hold a mix of things (a number here, a vector there), which is how most R results are returned.

3. Do something to every row, column, or item

R prefers you not to write loops. The apply family runs a function across a whole structure at once.

4. Writing your own function

When you repeat a calculation, wrap it in a function. The last line is what the function returns.

5. Reading your own data files

Most data arrives as a CSV or a tab-separated table. We host two small practice files for you, a gene-by-sample count table (practice-counts.csv) and a sample sheet (practice-samples.csv). The numbers are made up for practice, but the shape is exactly what a real RNA-seq experiment gives you. R can read them straight from the web:

For a file on your computer, give just its name. That works when the file sits in the folder R is looking in, its working directory:

getwd()                              # where is R looking right now?
counts <- read.csv("counts.csv")     # a CSV inside that folder
tsv    <- read.delim("data.tsv")     # tab-separated: read.delim()

6. tidyverse, going deeper: joins and reshaping

Two tidyverse moves you will use constantly: reshaping a wide table into the long form that ggplot likes, and joining two tables on a shared column. Order matters here: the count table has no sample column until you reshape it, so you reshape first, then join the sample sheet onto it.

library(tidyverse)
counts  <- read_csv("https://bionexushub.com/practice-counts.csv")
samples <- read_csv("https://bionexushub.com/practice-samples.csv")

# 1. wide (one column per sample) into long (one row per gene per sample)
long <- pivot_longer(counts, cols = -gene,
                     names_to = "sample", values_to = "count")

# 2. attach each sample's condition and batch, matched on "sample"
joined <- left_join(long, samples, by = "sample")
head(joined)

Try it in your browser (loads dplyr and tidyr the first time, so allow up to a minute):

7. Cleaner plots: facets and themes optional on first pass

Once your data is long and joined, a small plot per group is one line away with facet_wrap, and you can colour by the sample information you just joined.

ggplot(joined, aes(sample, count, fill = condition)) +
  geom_col() +
  facet_wrap(~ gene, scales = "free_y") +   # one mini-plot per gene, each with its own y-axis
  theme_minimal()

Try it: which genes go up in the treated samples? Then delete , scales = "free_y" and Run again to see why it is there.

For a deeper, opinionated take on making honest, readable figures, see Plots that tell the truth.

8. Bioconductor: where the biology packages live optional on first pass

Everyday packages come from CRAN (install.packages). Most bioinformatics packages live in a separate home called Bioconductor, installed through one helper.

install.packages("BiocManager")      # the gateway, installed once
BiocManager::install("DESeq2")        # then any Bioconductor package

Practice exercises

Type each into the scratchpad below or into RStudio, and run it. Peek at the solution only after you have tried.

Exercise 1: filter by condition. Using the named vector expr from section 1, keep only the genes with expression above 10.

expr[expr > 10]
# or, to get just the gene names:
names(expr[expr > 10])

Exercise 2: apply to each item. Use sapply to get the number of characters in each of c("ATG", "GG", "ATATAT").

sapply(c("ATG", "GG", "ATATAT"), nchar)
# 3 2 6

Exercise 3: a safe function. Write mean_or_zero(x) that returns the mean of x, or 0 when x is empty. New idea: if (condition) { A } else { B } runs A when the condition is TRUE and B otherwise, and length(x) tells you how many values x holds.

mean_or_zero <- function(x) {
  if (length(x) == 0) {
    0
  } else {
    mean(x)
  }
}

mean_or_zero(c(2, 4, 6))   # 4
mean_or_zero(c())             # 0

Check your understanding