Bulk RNA-seq: from FASTQ to differentially expressed genes

What you need

Before you start

• At least three biological replicates per condition. Fewer and the statistics have little to work with.
• A sample sheet whose condition names are exactly the ones you will compare.
• A note of anything that differs between samples apart from the condition: extraction day, sequencing run, operator. These are your batches.
• The reference for your organism, with the sequence and annotation from the same release.

Steps

1. 1

   Load your samples

   Uses Reads + Sample Sheet

   Drop a Reads step and a Sample Sheet step on the canvas and point them at your files. The sample sheet pairs each sample with its reads.

   Before going on, check every sample in the sheet has its files matched. A sample with no files, or a file with no sample, usually means a typo in a name.

2. 2

   Trim and check quality

   Uses fastp

   Connect the reads to fastp. It removes adapter sequence and low-quality bases, and writes a quality report per sample.

   Detect adapters (paired-end)

   On

   Finds the adapter from the read overlap, so you do not need to know the kit.

   Minimum length

   36

   Reads shorter than this after trimming map poorly and are dropped.

   Before going on, check the share of reads that pass the filter, which is usually above 90%. High duplication is normal in RNA-seq, because abundant transcripts are sequenced many times over.

3. 3

   Count reads per gene

   Uses Salmon

   Connect the trimmed reads and the reference to Salmon. It estimates how many reads came from each transcript and sums them to genes.

   Library type

   Automatic

   Salmon detects whether the library is stranded, which is easy to get wrong by hand.

   Index

   Transcriptome with genome decoys

   The decoys stop reads from unannotated regions being counted against a similar gene.

   Before going on, check the mapping rate per sample. Most well-annotated organisms map above 70%. A sample well below the others is worth a look before going further.

4. 4

   Test for differential expression

   Uses DESeq2

   Connect the counts and the sample sheet to DESeq2, tell it which condition to compare against which, and include your batches in the design.

   Design

   ~ batch + condition

   Accounts for batch before testing condition. Leave batch out and a batch difference can be read as biology.

   Contrast

   treated vs control

   The order matters: a positive fold change means higher in the first.

   Adjusted p-value

   0.05

   Corrected for testing every gene at once.

   Before going on, check the PCA before the gene list. Replicates should sit together, and the main split should be your condition. If it is the sequencing run instead, go back and add it to the design.

5. 5

   Draw the figures

   Uses Figures

   Bind a PCA, a volcano plot and a heatmap of the top genes to the DESeq2 output. They redraw whenever the step above them changes.

   Minimum |log2 fold change|

   1

   Draws the lines that mark a doubling or halving, so large but uncertain changes stand apart.

   Genes to label

   15

   Enough to orient a reader without covering the plot.

   Before going on, check that genes you expect not to change, such as common housekeeping genes, are not among the significant ones.

When it goes wrong

What you get

• A table of every gene with its fold change and adjusted p-value
• Normalised counts for every sample
• PCA, volcano plot and heatmap, exportable as SVG
• A quality report for every sample

Why it is done this way: Your first differential expression analysis, step by step.