1 RNA-Seq Expression Analysis

1 RNA-Seq Expression Analysis

1.1 Introduction
RNA sequencing (RNA-Seq) is a high-throughput method used to profile the transcriptome, quantify
gene expression and discover novel RNA molecules. This tutorial uses RNA sequencing of from two
Mycobacterium tuberculosis isolates (each with 3 replicates - 6 total samples) to walk you through
transcriptome pseudo-alignment and quantification and how to profile transcriptomic differences
between experimental conditions (Sensitive vs. Resistant) by identifying differentially expressed
genes.
For an introduction to RNA-Seq principles and best practices see:
A survey of best practices for RNA-Seq data analysis
Ana Conesa, Pedro Madrigal, Sonia Tarazona, David Gomez-Cabrero, Alejandra Cervera,
Andrew McPherson, Michał Wojciech Szcześniak, Daniel J. Gaffney, Laura L. Elo, Xue-
gong Zhang and Ali Mortazavi Genome Biol. 2016 Jan 26;17:13 doi:10.1186/s13059-
016-0881-8

1.2 Learning Outcomes

By the end of this practical, you should be familiar with:
• Using salmon to quantify transcript abundance
• Knowing how to create a study design file
• Loading the required files into R
• Creating a DESeq2 data set
• Running DESeq
• Visualising the relationship between samples
• Creating a results object
• Ranking genes by significance
• Visualisation of differential expression results
Focus on:
• Looking at what is in the files you are given, and the files you create
• How the information from different files relates and comes together
• Testing different inputs to commands to see what happens

1
1 RNA-Seq Expression Analysis 1.3 Practical Outline

1.3 Practical Outline

Workflow Overview. A schematic representation of each step in this practical exercise.

The main steps in all differential expression analyses are:

• Quantification - How many reads come from each gene? (Salmon)
• Normalisation - Dealing with biases in the data (DESeq2)
• Differential expression analysis (DESeq2)
• Visualisation and reporting (R libraries)
Additional Resources
RNA-seq workflow: gene-level exploratory analysis and differential expression

1.4 Authors
This tutorial was developed by Jon Ambler, Phelelani Mpangase and Nyasha Chambwe, based in
part, from materials from Victoria Offord and Adam Reid.

1.5 Prerequisites
This tutorial assumes that you have the following software or packages and their dependencies
installed on your computer. The software or packages used in this tutorial may be updated from
time to time so, we have also given you the version which was used when writing the tutorial. Where
necessary, instructions for how to download additional analysis tools are given in the relevant section.

Package Link for download/installation instructions Version

Trimmmomatic https://github.com/usadellab/Trimmomatic 0.39

2
 1 RNA-Seq Expression Analysis 1.6 Setup

Package Link for download/installation instructions Version

Salmon https://salmon.readthedocs.io/en/latest/index.html 1.4.0

R https://www.r-project.org/ 4.0.2
tximport https://bioconductor.org/packages/release/bioc/html/tximport.html
1.20.0
GenomicFeatures https://bioconductor.org/packages/release/bioc/html/GenomicFeatures.html
1.44.0
DESeq2 https://bioconductor.org/packages/release/bioc/html/DESeq2.html1.32
pheatmap https://cran.r-project.org/web/packages/pheatmap/index.html 1.0.12

1.6 Setup
1.6.1 Create Practical Directory
Navigate to the module folder on the Virtual Machine (VM) using the following command:
[ ]: cd /home/manager/course_data/rna_seq_pathogen

Create a working directory for the practical

[ ]: mkdir practical

Create a copy of the practial dataset to maintain the data integrity of the original dataset.
[ ]: cp /home/manager/course_data/rna_seq_pathogen/data/bacterial/* /home/manager/
,→course_data/rna_seq_pathogen/practical

Copy and paste the command above as a single line at your command line window.
Note: if you make a mistake at any point during this tutorial, you can reset by deleting the practical
folder and restarting from this section.
Move to the practical directory
[ ]: cd practical

1.6.2 Download Supplemental Datasets

Internet Access Required

Reference Transcriptome File Download the GFF formatted version of the annotation file.
Unfortunately due to the lack of standardization in bioinformatics, we need both the GTF file and the
GFF file for analysis (See the differences and similarities between these two file format specifications
here). Salmon uses the GTF file and the GenomicFeatures library in R needs a GFF file. Both files
contain the same annotation information, they are just formatted differently. No tools are able to
reliably convert one to the other because even GFF files are not formatted consistently.

3
 1 RNA-Seq Expression Analysis 1.6 Setup

[ ]: wget https://www.dropbox.com/s/4yjgbmy3dyhfoad/GCA_000195955.
,→2_ASM19595v2_genomic.gff?dl=1 -O /home/manager/course_data/rna_seq_pathogen/

,→practical/GCA_000195955.2_ASM19595v2_genomic.gff

Study Design File Download the study design file

The study design file is a tab separated file with 4 columns that encodes the phenotype and repeat
information on the samples we will analyze in this practical. You will import this file into R to setup
the differential expression analysis in the R Programming language. It is critical to keep track of the
details of which samples are which etc. for downstream analysis purposes.
[ ]: wget https://www.dropbox.com/s/6y3z9btz3bg20pn/practical_study_design.txt?dl=1␣
,→-O /home/manager/course_data/rna_seq_pathogen/practical/

,→practical_study_design.txt

Note: In this practical, we will use the existing study design file provided. However, in your own
work, you will have to create this file on your own. You can create this file in several ways some
of which could include exporting an Excel spreadsheet as a ’Tab delimited’ text file. You can also
use your favorite text editor.such as using the Text Editor app under the Applications menu on the
virtual machine.

RData Download “DE_data.RData” for later use.

These files will be used for differential expression analysis in a later section of this tutorial.
[ ]: wget https://www.dropbox.com/s/6onagsnpp9wrkrv/DE_data.RData.zip?dl=1 -O /home/
,→manager/course_data/rna_seq_pathogen/practical/DE_data.RData.zip

Uncompress the object:

4
 2 Introducing The Tutorial Dataset

[ ]: unzip DE_data.RData.zip

1.6.3 Setup R Analysis Environment

Internet Access Required
In this section you will download additional R Libraries, packages of analysis tools to support dif-
ferential expression analysis in R.
Start R by typing the following command:
[ ]: R

The Bioconductor Project provides tools written in R for the ’analysis and comprehension of high-
throughput genomic data’. Here we will install two additional software packages (GenomicFeatures
and tximport) for our practical today (See additional guidelines for Bioconductor package installa-
tion).
[ ]: # Bioconductor Packages
install.packages("BiocManager")
BiocManager::install("GenomicFeatures", force = TRUE)
BiocManager::install("tximport")

# CRAN Packages
install.packages("pheatmap")

When asked whether to:

Update all/some/none? [a/s/n]
Select no (n)
Exit R.
[ ]: q()

2 Introducing The Tutorial Dataset

Working through this tutorial, you will investigate the differences in expression between two My-
cobacterium tuberculosis isolates with different phenotypes (Sensitive vs. Resistant). You will
analyze 6 samples across the two conditions. The experiment is setup in the following way:

run unique_id phenotype repeat

N2 sample1 Resistant 1
N6 sample2 Resistant 2
N10 sample3 Resistant 3
N14 sample4 Sensitive 1

5
 3 Estimate Transcript Abundance With Salmon 2.1 Exercise 1

run unique_id phenotype repeat

N18 sample5 Sensitive 2

N22 sample6 Sensitive 3

Research Question: what genes are differentially expressed between these two isolates
that can explain the differences in their phenotypes?
Check that you can see the FASTQ files in the practical directory.
[ ]: ls N*.fq.gz

The FASTQ files contain the raw sequence reads for each sample. There will typically be four lines
per read:
1. Header
2. Sequence
3. Separator (usually a ’+’)
4. Encoded quality value
Take a look at one of the FASTQ files.
[ ]: zless N2_sub_R1.fq.gz | head

You can find out more about the FASTQ format at https://en.wikipedia.org/wiki/FASTQ_
format.

2.1 Exercise 1
2.2 Questions
Q1: Why is there more than one FASTQ file per sample?
Hint: think about why there is a N2_sub_R1.fq.gz and a N2_sub_2.fq.gz
Q2: How many reads were generated for the N2 sample?
Hint: we want the total number of reads from both files (N2_sub_R1.fq.gz and N2_sub_2.fq.gz) so
perhaps think about the FASTQ format and the number of lines for each read or whether there’s anything
you can use in the FASTQ header to search and count.
Q3: The three Resistant samples N2, N6 and N10 represent technical replicates. True or False?
Comment on your answer

3 Estimate Transcript Abundance With Salmon

In this section, you will use Salmon, a transcript quantification method to estimate the number of
reads in each sample that map to reference transcripts. This count is an estimate of the abundance
or expression level of these transcripts in our experiment. As Salmon is an alignment free method,
read quantification does not require an input BAM file. Salmon using a selective mapping algorithm

6
 3 Estimate Transcript Abundance With Salmon 3.1 Create Transcriptome Index

to align the reads directly to a set of target transcripts such as those from a reference database for
your organism.
Inputs include: * A set of target transcripts - FASTA format Reference Transcriptome
GCA_000195955.2_ASM19595v2_genomic.transcripts.fa * Sample reads - FASTA/FASTQ files for
your sample N2_sub_R1.fq.gz.....
See more details in:
Reference:
Patro R, Duggal G, Love MI, Irizarry RA, Kingsford C. Salmon provides fast and bias-
aware quantification of transcript expression. Nat Methods. 2017 Apr;14(4):417-
419. doi: 10.1038/nmeth.4197. PMID: 28263959; PMCID: PMC5600148.

3.1 Create Transcriptome Index

First, we create the necessary index files for any alignment tools downstream. The salmon index
is created from the transcripts file, and while it is named transcripts_index here, you can name it
anything.
[ ]: salmon index -t GCA_000195955.2_ASM19595v2_genomic.transcripts.fa \
-i transcripts_index -k 31

Take a look at the various files created in the transcripts_index folder created by salmon.

3.2 Transcript Quantification

Next we perform quantification with salmon using the transcript index folder we just created.
However, it is important to perform key quality control analysis of your sample reads (*.fq.gz) files
before proceeding with quantification.

3.2.1 Read Quality Control Analysis

Here we demonstrate how to trim and filter the reads using Trimmomatic. Here we clean the reads
by looking at various quality metrics including:
• low quality reads - filter based on a min leading base quality of 3, and trailing base quality
of 3. The scan with a sliding window that cuts when the average quality is lower than 15.
• Illumina adapter sequences - clip out these technical artifacts using TruSeq3-PE.fa
• read length - discard any ready with a length lower than 36
[ ]: trimmomatic PE N2_sub_R1.fq.gz N2_sub_R2.fq.gz \
N2_trimmed_forward_paired.fq.gz N2_trimmed_forward_unpaired.fq.gz \
N2_trimmed_reverse_paired.fq.gz N2_trimmed_reverse_unpaired.fq.gz \
ILLUMINACLIP:/home/manager/miniconda/pkgs/trimmomatic-0.39-1/share/
,→trimmomatic-0.39-1/adapters/TruSeq3-PE.fa:2:30:10:2:keepBothReads \

LEADING:3 TRAILING:3 MINLEN:36

In a typically workflow, you would do fastQC analysis before and after to check the effects of the
trimming, but as that is not the purpose of this section, we will skip that for now.

7
 3 Estimate Transcript Abundance With Salmon 3.2 Transcript Quantification

3.2.2 Transcript Quantificaiton using salmon

Now we used the trimmed and paired reads to estimate transcript abundance:
[ ]: salmon quant \
--geneMap GCA_000195955.2_ASM19595v2_genomic.gtf \
--threads 2 -l A \
-i transcripts_index GCA_000195955.2_ASM19595v2_genomic.fa \
-1 N2_trimmed_forward_paired.fq.gz \
-2 N2_trimmed_reverse_paired.fq.gz -o N2

This creates a new folder named “N2” in the directory, which contains a number of files:

The “quant” files are the files containing the read counts for the genes. These are what downstream
tools like DESeq2 or edgeR would use for the differential expression analysis.
It is not always practical to run commands for each sample one at a time. So we can write small
scripts to process multiple samples using the same set of commands as is shown here:
[ ]: for r1 in *_R1.fq.gz
do
echo $r1
sample=$(basename $r1)
sample=${sample%_sub_R1.fq.gz}
echo "Processing sample: "$sample

trimmomatic PE ${sample}_sub_R1.fq.gz ${sample}_sub_R2.fq.gz \

${sample}_trimmed_forward_paired.fq.gz ${sample}_trimmed_forward_unpaired.fq.
,→gz \

${sample}_trimmed_reverse_paired.fq.gz ${sample}_trimmed_reverse_unpaired.fq.
,→gz \

ILLUMINACLIP:/home/manager/miniconda/pkgs/trimmomatic-0.39-1/share/
,→trimmomatic-0.39-1/adapters/TruSeq3-PE.fa:2:30:10:2:keepBothReads \

LEADING:3 TRAILING:3 MINLEN:36

salmon quant --geneMap GCA_000195955.2_ASM19595v2_genomic.gtf \

--threads 2 -l A -i transcripts_index GCA_000195955.2_ASM19595v2_genomic.fa \

8
 4 Differential Expression Analysis with DESeq2 3.3 Questions

-1 ${sample}_trimmed_forward_paired.fq.gz \
-2 ${sample}_trimmed_reverse_paired.fq.gz \
-o ${sample}

done

3.3 Questions
In the N2 folder, take a look at the quant.sf file and answer the following questions:
• Q1: What does TPM stand for?
• Q2: How many reads in total are mapped to the tRNA genes? (They start with “rna-”)
• Q3: Do you think this level of tRNA is acceptable?
Take a moment to discuss what the effect of large amounts of reads from tRNA and rRNA could
have on normalisation.
Hint: For further help understanding the quant.sf* output files look at this salmon documentation
page.*

4 Differential Expression Analysis with DESeq2

In this section, you will use the R Package DESeq2 to perform differential expression analysis using
the salmon quantification files you generated in the previous section. For more information about
DESeq can look at the manuscript describing the method:
Reference:
Love MI, Huber W, Anders S (2014). “Moderated estimation of fold change and
dispersion for RNA-seq data with DESeq2.” Genome Biology, 15, 550. doi:
10.1186/s13059-014-0550-8.
We will now move to working in the R Programming environment: ## Load Data into the R Envi-
ronment
Start R
[ ]: R

Load the required libraries

R has a base set of classes and methods and tools. The additional packages we installed are a set
of functions and tools designed to handle specific biological data types and support analyses and
visualization such as of RNA-seq data. Here, we load those software packages that are relevant to
our analysis.
[ ]: library("tximportData")
library("tximport")
library("GenomicFeatures")
library("DESeq2")

9
 4 Differential Expression Analysis with DESeq2

library("pheatmap")

Load the study design table

[ ]: design_file <-"practical_study_design.txt"

samples <- read.table(design_file, header=TRUE)

# Set the row names as the samples names

rownames(samples) <- samples$run

samples

If we look at the samples object, we can see the data from our study design file.
Link Salmon Output File Paths to Samples
Create a files object pointing to the related quant.sf file for each sample
[ ]: root_dir <- "/home/manager/course_data/rna_seq_pathogen/practical"
files <- file.path(root_dir, samples$run, "quant.sf")
files

In the files object, we should now see the path to the quant.sf file in each sample’s folder. This is
much simpler and less error-prone than typing out each file path manually.
Use the run column to map samples to their file paths
[ ]: names(files) <- samples$run

The files object now has the sample name linked to the file path and can be used as a way to map
the information.

4.0.1 Make Annotation Database Object

Load the annotation information
[ ]: path_to_gff <- "GCA_000195955.2_ASM19595v2_genomic.gff"
txdb <-makeTxDbFromGFF(path_to_gff, organism="Mycobacterium tuberculosis")

The makeTxDbFromGFF function is part of the GenomicFeatures library and is used to create a
Transcript Database (txdb) object from a GFF annotation file. The TxDb class is a container for
storing transcript annotations.
Extract the GENEID key values from the tbdx object
[ ]: k <- keys(txdb, keytype = "GENEID")
head(k)

The k object is a list of all the gene names based on the GENEID as extracted from the annotation
file.

10
 4 Differential Expression Analysis with DESeq2

Create a mapping table based on the GENEID and TXNAME info

[ ]: tx2gene <- select(txdb, keys = k, keytype = "GENEID", columns = "TXNAME")
head(tx2gene)

Take a look at the output, you will see 2 columns. This will be used to map the gene names from
the salmon files to the annotation files.
Rename the genes in the GENEID column
We do this to match the gene names found in the salmon quant.sf files with those found in the
transcriptome and annotation files.
[ ]: tx2gene[["GENEID"]] <- with(tx2gene, ifelse(!grepl("rna-", TXNAME),␣
,→paste0("gene-", TXNAME), TXNAME))

head(tx2gene[["GENEID"]])

The above command goes through the tx2gene object, and looks for where the gene name in the
TXNAME column DOES NOT have the prefix ”rna-”. In those rows, it adds the ”gene-” prefix to
the gene name in the GENEID column.
When the annotation file is imported by makeTxDbFromGFF , it removes the “gene-” prefix that
is present in the quant files. This is because of the way that the annotation and transcript files are
formatted.
You will see in the GFF and transcript files, that the gene ID is formatted like this:
ID=gene-Rv0005
Whereas in the GTF file it is formatted like this:
gene_id ”Rv0005”;
Though salmon uses the GTF file, it adds on the gene- prefix where it is missing while
makeTxDbFromGFF removes it where it is present.
[ ]: txi <- tximport(files, type="salmon", tx2gene=tx2gene)

4.0.2 Create the DESeq data set object

[ ]: ddsTxi <- DESeqDataSetFromTximport(txi, colData = samples, design = ~ phenotype)

The inputs to create a DESeq object are:

• txi - the summary of transcript level abundance estimates
• colData - information from the study design file loaded earlier
• design - formula that expresses the relationship between the gene counts and the variables in
the study design. Here, we are using phenotype.
The design formula is a statement that specifies how we want to model the variation in gene ex-
pression from the abundance estimates for each sample (counts). This statement is loosely saying
that we expect that the level of expression of a gene is dependent on the phenotype of the bacterial

11
 4 Differential Expression Analysis with DESeq2 4.1 Exploratory Visualization

isolate. If we have more than a single experimental factor to consider, we would change how we
specify the design formula.
Note: For a more extensive treatment of how to setup the design formula for more complex experi-
mental designs, read through A guide to creating design matrices for gene expression experiments.

4.0.3 Perform DESeq Analysis

Normalisation and model fitting with DESeq2
Next we used DESeq to normalize the data and fit a model that relates the gene count information
to the phenotype.
[ ]: ddsTxi <- DESeq(ddsTxi)

This step does the actual normalisation and model fitting for the data, and results in a deseq data
set.
Load provided .RData for differential expression analysis:
[ ]: load("DE_data.RData")

4.1 Exploratory Visualization

Transform data for visualization:
The rlog transformation provides functionality for visualization and clustering.
[ ]: # R log norm
rldTxi <- rlog(ddsTxi, blind = FALSE)

Plot principal component analysis (PCA):

The PCA plot allow us to assess the similarities and differences between the study samples in order
to determine if the data fit our expectation compared to the experimental design. This is a good
quality control check to use to ensure that we did not introduce any errors during sample processing,
sequencing and primary data analysis steps.
[ ]: # Plot PCA
plotPCA(rldTxi, intgroup = c("phenotype"))

4.2 Result Generation and Exploration

Get summary of differentially expression results:
The results function of DESeq2 is used to get the the results of the DESeq function ranked by
metrics to use to determine which genes are significantly differentially expressed.
[ ]: summary(results(ddsTxi))

Apply LFC threshold and adjusted p-value cutoffs to get significantly differentially expressed
genes:

12
 5 Key Aspects of Differential Expression Analysis 4.3 Questions

A significance threshold (FDR, or alpha) and log2FC threshold can be passed to the results function
to filter non-signficant differentially expressed genes.
[ ]: resTxi <- results(ddsTxi, alpha=0.05, lfcThreshold=0.5,␣
,→altHypothesis="greaterAbs")

summary(resTxi)

Get significantly differentially expressed genes:

[ ]: sigTxi <- resTxi[!is.na(resTxi$padj) & resTxi$padj < 0.05 &␣
,→abs(resTxi$log2FoldChange) >= 0.5,]

sigTxi

Visualize your differentially expressed genes:

Extract the rlog transformed expression values for the significant gene to use for the heatmap.
[ ]: matrixTxi <- assay(rldTxi)[rownames(sigTxi), ]
matrixTxi

Plot heatmap:
[ ]: pheatmap(matrixTxi, cluster_rows=FALSE, show_rownames=TRUE, cluster_cols=TRUE)

4.3 Questions
Q1. How many genes are up- and down-regulated between the resistant and sensitive isolates
of the Mycobacterium tuberculosis?
Hint:use the summary function on the DESeq results object.
Q2: How many genes are significantly differentially expressed (i.e., meet the LFC threshold
and adjusted p-value cutoffs) between the resistant and sensitive isolates of the Mycobac-
terium tuberculosis? Name these genes.
Q3. What are the p-values for the significantly differentially expressed genes?

5 Key Aspects of Differential Expression Analysis

5.1 Replicates and power
’Biological replicates are parallel measurements of biologically distinct samples that
capture random biological variation, which may itself be a subject of study or a noise
source.
Technical replicates are repeated measurements of the same sample that represent in-
dependent measures of the random noise associated with protocols or equipment’
Blainey, Paul et al. “Points of significance: replication.” Nature methods vol. 11,9
(2014): 879-80. doi:10.1038/nmeth.3091

13
5 Key Aspects of Differential Expression Analysis 5.2 p-values vs. q-values

In order to accurately ascertain which genes are differentially expressed and by how much it is
necessary to use replicated data. As with all biological experiments doing it once is simply not
enough. There is no simple way to decide how many replicates to do, it is usually a compromise
between statistical power and cost. By determining how much variability there is in the sample
preparation and sequencing reactions, we can better assess how highly genes are really expressed
and more accurately determine any differences. The key to this is performing biological rather than
technical replicates. This means, for instance, growing up three batches of parasites, treating them
all identically, extracting RNA from each and sequencing the three samples separately. Technical
replicates, whereby the same sample is sequenced three times do not account for the variability that
really exists in biological systems or the experimental error between batches of parasites and RNA
extractions.
Note: more replicates will help improve power for genes that are already detected at high levels, while
deeper sequencing will improve power to detect differential expression for genes which are expressed at
low levels.

5.2 p-values vs. q-values

When asking whether a gene is differentially expressed we use statistical tests to assign a p-value. If
a gene has a p-value of 0.05, we say that there is only a 5% chance that it is not really differentially
expressed. However, if we are asking this question for every gene in the genome, then we would
expect to see p-values less than 0.05 for many genes even though they are not really differentially
expressed. Due to this statistical problem, we must correct the p-values so that we are not tricked
into accepting a large number of erroneous results. Q-values are p-values which have been corrected
for what is known as multiple hypothesis testing. Therefore, it is a qvalue of less than 0.05 that we
should be looking for when asking whether a gene is signficantly differentially expressed.

5.3 Altermate software

If you have a good quality genome and genome annotation such as for model organisms e.g. human,
mouse, Plasmodium etc. map to the transcriptome to determine transcript abundance. This is even
more relevant if you have variant transcripts per gene as you need a tool which will do its best to
determine which transcript is really expressed. Kallisto (Bray et al. 2016; PMID: 27043002 and
eXpress (Roberts & Pachter, 2012; PMID: 23160280 are examples of these tools.

5.4 What do I do with a list of differentially expressed genes?

Differential expression analysis results are a list of genes which show differences between conditions.
It can be daunting trying to determine what the results mean. On one hand, you may find that there
are no real differences in your experiment. Is this due to biological reality or noisy data? On the
other hand, you may find several thousands of genes are differentially expressed.
What can you say about that?
Other than looking for genes you expect to be different or unchanged, one of the first things to
do is look at common themes in gene functionality across your list. For example, you can carry
out Gene Ontology (GO) term enrichment analysis. There are many different algorithms for this,
but you could annotate your genes with functional terms from GO using for instance Blast2GO
(Conesa et al., 2005; PMID: 16081474) and then use TopGO (Alexa et al., 2005; PMID: 16606683)

14
6 Normalisation

to determine whether any particular sorts of genes occur more than expected in your differentially
expressed genes.

6 Normalisation
6.1 Introduction
In a previous section, we looked at estimating transcript abundance with Salmon. The abundances
are reported as transcripts per million (TPM), but what does TPM mean and how is it calculated?
The objectives of this part of the tutorial are:
• understand why RNA-Seq normalisation metrics are used
• understand the difference between RPKM, FPKM and TPM
• understand how RPKM and TPM are calculated for a gene of interest
There are many useful websites, publications and blog posts which go into much more detail about
RNA-Seq normalisation methods. Here are just a couple (in no particular order):
• What the FPKM? A review of RNA-Seq expression units
• RPKM, FPKM and TPM, clearly explained
• A survey of best practices for RNA-seq data analysis
• The RNA-seq abundance zoo

6.2 Why do we use normalisation units instead of raw counts?

Raw reads counts are the number of reads originating from each transcript which can be affected
by several factors:
• sequencing depth (total number of reads)
The more we sequence a sample, the more reads we expect to be assigned.
• gene/transcript length
The longer the gene or transcript, the more reads we expect to be assigned to it.

15
6 Normalisation 6.2 Why do we use normalisation units instead of raw counts?

Figure 4. Effect of sequencing depth and gene length on raw read counts

Look at the top part of Figure 4. In which sample, X or Y, is the gene more highly expressed?
Neither, it’s the same in both. What we didn’t tell you was that the total number of reads generated
for sample A was twice the number than for sample B. That meant almost twice the number of reads
are assigned to the same gene in sample A than in sample B.
Look at the bottom part of Figure 4. Which gene, X or Y, has the greatest gene level expression?
Neither, they are both expressed at the same level. This time we didn’t tell you that gene X is twice
the length of gene Y. This meant that almost twice the number reads were assigned to gene X than
gene Y.
In the top part of Figure 4, the gene in sample X has twice the number of reads assigned to it than the
same gene in sample Y. What isn’t shown is that sample X had twice the number or total reads than
sample Y so we would expect more reads to be assigned in sample X. Thus, the gene is expressed at
roughly the same level in both samples. In the bottom part of Figure 4, gene X has twice the number
of reads assigned to it than gene Y. However, gene X is twice the length of gene Y and so we expect
more reads to be assigned to gene X. Again, the expression level is roughly the same.

6.2.1 Reads per kilobase per million (RPKM)

Reads per kilobase (of exon) per million (reads mapped) or RPKM is a within sample normalisation
method which takes into account sequencing depth and length biases.
To calculate RPKM, you first normalise by sequencing depth and then by gene/transcript length.
1. Get your per million scaling factor
Count up the total number of reads which have been assigned (mapped) in the sample. Divide
this number by 1,000,000 (1 million) to get your per million scaling factor (N).
2. Normalise for sequencing depth
Divide the number of reads which have been assigned to the gene or transcript (C) by the per
million scaling factor you calculated in step 1. This will give you your reads per million (RPM).

16
6 Normalisation 6.2 Why do we use normalisation units instead of raw counts?

3. Get your per kilobase scaling factor

Divide the total length of the exons in your transcript or gene in base pairs by 1,000 (1 thou-
sand) to get your per kilobase scaling factor (L).
4. Normalise for length
Divide your RPM value from step 2 by your per kilobase scaling factor (length of the gene/-
transcript in kilobases) from step 3. This will give you your reads per kilobase per million or
RPKM.
This can be simplified into the following equation:

C
RP KM =
LN

Where:
• C is number of reads mapped to the transcript or gene
• L is the total exon length of the transcript or gene in kilobases
• N is the total number of reads mapped in millions

6.2.2 Fragments per kilobase per million (FPKM)

Fragments per kilobase per million or FPKM is essentially the same as RPKM except that:
• RPKM is designed for single-end RNA-Seq experiments
• FPKM is designed for paired-end RNA-Seq experiments
In a paired-end RNA-Seq experiment, two reads may be assigned to a single fragment (in any ori-
entation). Also, in some cases, only one of those reads will be assigned to a fragment (singleton).
The only difference between RPKM and FPKM is that FPKM takes into consideration that two reads
may be assigned to the same fragment.

6.2.3 Transcripts per million (TPM)

Calculating the transcripts per million or TPM is a similar process to RPKM and FPKM. The main
difference is that you will first normalise for length bias and then for sequencing depth bias. In a
nut shell, we are swapping the order of normalisations.
1. Get your per kilobase scaling factor
Divide the total length of the exons in your transcript in base pairs by 1,000 (1 thousand) to
get your per kilobase scaling factor.
2. Normalise for length
Divide the number of reads which have been assigned to the transcript by the per kilobase
scaling factor you calculated in step 1. This will give you your reads per kilobase (RPK).
3. Get the sum of all RPK values in your sample
Calculate the RPK value for all of the transcripts in your sample. Add all of these together to
get your total RPK value.

17
6 Normalisation 6.3 Calculating RPKM and TPM values

4. Get your per million scaling factor

Divide your total RPK value from step 3 by 1,000,000 (1 million) to get your per million scaling
factor.
5. Normalise for sequencing depth
Divide your RPK value calculated in step 2 by the per million scaling factor from step 4. You
now have your transcripts per millions value or TPM.

6.3 Calculating RPKM and TPM values

To try and answer this, let’s look at a worked example. Here, we have three genes (A-C) and three
biological replicates (1-3).

Gene Length Replicate 1 Replicate 2 Replicate 3

A 2,000 bases 10 12 30
B 4,000 bases 20 25 60
C 1,000 bases 5 8 15

There are two things to notice in our dataset:

• Gene B has twice number reads mapped than gene A, possibly as it’s twice the length
• Replicate 3 has more reads mapped than any of the other replicates, regardless of which gene
we look at

6.3.1 Calculating RPKM

Step 1: get your per million scaling factor In the table below is the total number of reads which
mapped for each of the replicates. To get our per million scaling factor, we divide each of these
values by 1,000,000 (1 million).

Gene Replicate 1 Replicate 2 Replicate 3

Total reads mapped 3,500,000 4,500,000 10,600,000

Per million reads 3.5 4.5 10.6

Step 2: normalise for sequencing depth We now divide our read counts by the per million scaling
factor to get our reads per million (RPM).
Before:

18
6 Normalisation 6.3 Calculating RPKM and TPM values

Gene Replicate 1 Replicate 2 Replicate 3

A 10 12 30
B 20 25 60
C 5 8 15

After:

Gene Replicate 1 RPM Replicate 2 RPM Replicate 3 RPM

A 2.857 2.667 2.830

B 5.714 5.556 5.660
C 1.429 1.778 1.415

Step 3: get your per kilobase scaling factor Here we have our gene length in base pairs. For our
per kilobase scaling factor we need to get our gene length in kilobases by dividing it by 1,000.

Gene Length (base pairs) Length (kilobases)

A 2,000 2
B 4,000 4
C 1,000 1

Step 4: normalise for length Finally, we divide our RPM values from step 2 by our per kilobase
scaling factor from step 3 to get our reads per kilobase per million (RPKM).
Before:

Gene Replicate 1 RPM Replicate 2 RPM Replicate 3 RPM

A 2.857 2.667 2.830

B 5.714 5.556 5.660
C 1.429 1.778 1.415

After:

19
6 Normalisation 6.3 Calculating RPKM and TPM values

Gene Replicate 1 RPKM Replicate 2 RPKM Replicate 3 RPKM

A 1.43 1.33 1.42

B 1.43 1.39 1.42
C 1.43 1.78 1.42

Notice that even though replicate 3 had more reads assigned than the other samples and a greater
sequencing depth, its RPKM is quite similar. And, that although gene B had twice the number of
reads assigned than gene A, its RPKM is the same. This is because we have normalised by both
length and sequencing depth.

6.3.2 Calculating TPM

Now we’re going to calculate the TPM values for the same example data. As a reminder, here are
our three genes (A-C) and three biological replicates (1-3).

Gene Length Replicate 1 Replicate 2 Replicate 3

A 2,000 bases 10 12 30
B 4,000 bases 20 25 60
C 1,000 bases 5 8 15

Step 1: get your per kilobase scaling factor Again, our gene lengths are in base pairs. For our
per kilobase scaling factor we need to get our gene length in kilobases by dividing it by 1,000.

Gene Length (base pairs) Length (kilobases)

A 2,000 2
B 4,000 4
C 1,000 1

Step 2: normalise for length Now we divide the number of reads which have been assigned to
each gene by the per kilobase scaling factor we just calculated. This will give us our reads per
kilobase (RPK).
Before:

Gene Replicate 1 Replicate 2 Replicate 3

A 10 12 30
B 20 25 60

20
6 Normalisation 6.3 Calculating RPKM and TPM values

Gene Replicate 1 Replicate 2 Replicate 3

C 5 8 15

After:

Gene Replicate 1 Replicate 2 Replicate 3

A 5 6 15
B 5 6.25 15
C 5 8 15

Step 3: get the sum of all RPK values in your sample Next, we sum the RPK values for each of
our replices. This will give use our total RPK value for each replicate. To make this example scalable,
we assume there are other genes so the total RPK is made up.

Gene Replicate 1 Replicate 2 Replicate 3

A 5 6 15
B 5 6.25 15
C 5 8 15
... ... ... ...
Total RPK 150,000 202,500 450,000

Step 4: get your per million scaling factor Here, instead of dividing our total mapped reads
by 1,000,000 (1 million) to get our per million scaling factor, we divide our total RPK values by
1,000,000 (1 million).

Gene Replicate 1 Replicate 2 Replicate 3

Total RPK 150,000 202,500 450,000

Per million RPK 0.1500 0.2025 0.4500

Step 5: normalise for sequencing depth Finally, we divide our individual RPK values from step
2 by the per million scaling factor in step 4 to give us our TPM values.
Before:

21
6 Normalisation 6.4 Which normalisation unit should I use?

Gene Replicate 1 Replicate 2 Replicate 3

A 5 6 15
B 5 6.25 15
C 5 8 15

After:

Gene Replicate 1 Replicate 2 Replicate 3

A 33.33 29.63 33.33

B 33.33 30.86 33.33
C 33.33 39.51 33.33

6.4 Which normalisation unit should I use?

Well, there’s a lot of debate around this, so let’s look at our total normalised values for each replicate.

6.4.1 RPKM

Gene Replicate 1 RPKM Replicate 2 RPKM Replicate 3 RPKM

A 1.43 1.33 1.42

B 1.43 1.39 1.42
C 1.43 1.78 1.42
Total RPKM 4.29 4.50 4.25

6.4.2 TPM

Gene Replicate 1 Replicate 2 Replicate 3

A 33.33 29.63 33.33

B 33.33 30.86 33.33
C 33.33 39.51 33.33
Total TPM 100 100 100

22
6 Normalisation 6.4 Which normalisation unit should I use?

Notice that that total TPM value for each of the replicates is the same. This is not true for RPKM
and FPKM where the total values differ. With TPM, having the same total value for each replicate
makes it easier to compare the proportion of reads mapping to each gene across replicates (although
you shouldn’t really compare across experiments). With RPKM and FPKM, the differing total values
make it much harder to compare replicates.

Congratulations, you have reached the end of this tutorial!

23