Warning
This page is located in archive.

Tutorial 9 - RNA-Seq - alignment, abundance quantification and differential expression analysis

Introduction

Analysis of differential gene expression is one of the most popular bioinformatics tasks. Each particular gene can be either turned on or off. This process regulates the amount of produced proteins in each cell, allowing it to react to a lack of minerals or change its type from stem cell to neuron. It is, however, problematic to measure protein levels directly. Therefore scientists developed a method to use DNA sequencing to read RNA transcripts. The cell is killed, and all RNA is put into a sequencing machine. Machine reads (again, only short fragments!) this RNA. To quantify the abundance of each gene, we must first map reads to each particular gene.

The mapping is not as easy as it seems. Firstly, we need to have a reference genome with its annotation with the boundaries of all genes. Secondly, not all organisms are the same; there may be some differences between different representatives of a single species. Therefore, reads won't match exactly. Not to mention any sequencing errors and alternative splicing, which means that reads won't match to a single interval of a genome but two or more consecutive intervals. Repeats and gene-similarity produce another source of uncertainty in the mapping as we cannot be sure which gene the transcript originated from. Mapping algorithms (as STAR) need first to create an index, which is a suffix tree on the reference genome to be able to search in it. Multiple gene mapping reads can be solved, for example, by a procedure similar to EM-algorithm. Secondly, the mapping produces aligned reads in a BAM file.

Having the mapping, we can use a program called htseq-count to calculate a vector with counts for each gene. The counts need to be normalized as longer genes mean more reads with the same number of transcripts. Those counts are then direct input to the DeSEQ2 library, which allows different statistical analyses. If you have multiple samples, you may compare which genes are “more active” in cancer cells compared to healthy cells and use this knowledge further. In the next tutorial, we will try ourselves to run the mapping algorithm, estimate the counts, and use them in a simple analysis of zebrafish embryos. The problem is simplified so that we can run it on a modest work-station in a period dedicated to one tutorial. Therefore, only a single gene shows statistically significant differences in the single chromosome we use.

Before the tutorial

As we will work with bulk data, download some files in advance. This following script should do all the required for you.

bash

mkdir bintut9
cd bintut9

# annotation files and genome
wget ftp://ftp.ensembl.org/pub/release-96/gtf/danio_rerio/Danio_rerio.GRCz11.96.gtf.gz
wget ftp://ftp.ensembl.org/pub/release-96/fasta/danio_rerio/dna/Danio_rerio.GRCz11.dna.chromosome.1.fa.gz

gunzip Danio_rerio.GRCz11.dna.chromosome.1.fa.gz
gunzip Danio_rerio.GRCz11.96.gtf.gz

wget ftp://ftp.ebi.ac.uk/pub/training/Train_online/RNA-seq_exercise/6h_1.fastq
wget ftp://ftp.ebi.ac.uk/pub/training/Train_online/RNA-seq_exercise/6h_2.fastq

wget ftp://ftp.ebi.ac.uk/pub/training/Train_online/RNA-seq_exercise/2cells_1.fastq
wget ftp://ftp.ebi.ac.uk/pub/training/Train_online/RNA-seq_exercise/2cells_2.fastq

wget https://github.com/alexdobin/STAR/archive/2.7.0f.tar.gz
tar -xzf 2.7.0f.tar.gz

wget https://www.ebi.ac.uk/~emily/Online%20courses/NGS/RNA-seq.zip
unzip RNA-seq.zip
mkdir tophat
mv RNA-seq/annotation/Danio_rerio.Zv9.66.gtf tophat/
mv RNA-seq/tophat/ZV9_6h/accepted_hits.sorted.bam tophat/6h.bam
mv RNA-seq/tophat/ZV9_2cells/accepted_hits.sorted.bam tophat/2cells.bam

Next, we need to install the DESeq2 library into R. Run the following code in your R installation.

R

source('http://bioconductor.org/biocLite.R')
biocLite('DESeq2')
library(DESeq2)

Introduction

In the tutorial, we will study expression levels of embryo cells of zebrafish, Danio rerio in Latin. Sequencing was done with Illumina RNA-seq platform which produces reads of RNA transcripts in a cell.

You have downloaded the following files:

  1. Danio_rerio.GRCz11.dna.chromosome.1.fa.gz with the DNA sequence of the zebrafish. Due to the time constraints in class, we will use only chromosome 1 in our experiments.
  2. Danio_rerio.GRCz11.96.gtf.gz with annotation of the genome. The file describes where each gene is located.
  3. Files 2cells_1.fastq and 2cells_2.fastq contain paired-end RNA-seq reads of a two-cell zebrafish embryo.
  4. Files 6h_1.fastq and 6h_2.fastq contain paired-end RNA-seq reads of zebrafish embryo $6$ hours post fertilization.
  5. 2.7.0f.tar.gz contains the STAR aligner, see https://doi.org/10.1093/bioinformatics/bts635.
  6. RNA-seq.zip contains aligned reads by tophat aligner. For time constraints, we are not able to align reads in the class.

The reference genome and annotation may be downloaded from the ENSEMBL database.

The input data were obtained from the experiment described here and here. The sample alignment was downloaded from this tutorial.

Alignment

The first step in the pipeline is the alignment of the reads to the genome. We have downloaded the annotated references from the ENSEMBL database. For the alignment, we will use the STAR aligner to calculate the alignment. The original paper was published here:

  1. Alexander Dobin, Carrie A. Davis, Felix Schlesinger, Jorg Drenkow, Chris Zaleski, Sonali Jha, Philippe Batut, Mark Chaisson, Thomas R. Gingeras, STAR: ultrafast universal RNA-seq aligner, Bioinformatics, Volume 29, Issue 1, January 2013, Pages 15–21, https://doi.org/10.1093/bioinformatics/bts635

This aligner is faster than the tophat aligner that was used commonly before; STAR, however, consumes a lot of memory. Aligning human reads requires approximately $32\,\mathrm{GB}$. Hence, we will use only the first chromosome in the class.

The alignment has two steps. First, we need to create the index to calculate the alignment. Use the following parameters:

  1. The number of threads should be equal to the number of cores of your computer.
  2. Create a directory genome, and use this directory for the genome index.
  3. The genome FASTA file is the file with extension .fa
  4. Use the .gtf file you downloaded.
  5. Sjdb overhang should be equal to read length - 1. Use zless to view one of the read files and count the nucleotides.
  6. Don't forget to set the program mode.

Use documentation of the aligner to set the parameters: https://github.com/alexdobin/STAR/blob/master/doc/STARmanual.pdf

The second step is the alignment itself. The parameters are as follows.

  1. The number of threads should be equal to the number of cores.
  2. Use the genome directory you created in the last step.
  3. The input should contain two FASTQ files for a single condition. File _1.fastq contains forward reads, file _2.fastq contains backward reads.
  4. Add outSAMtype to sort the entries by coordinate in the final BAM file.
  5. Set prefix of the output as starout/6h or starout/2cells.
Unfortunately, the alignment takes around half an hour on a reasonably fast computer. Therefore, in the next step, we will use precomputed aligned files.

The produced BAM files are compressed versions of SAM file format. Feel free to inspect the contents of the output. You would need samtools to do that.

Abundance quantification

For abundance quantification, we will use the HTSeq count python tool. This tool is aimed as replacement of the cufflinks alternative. Documentation of the tool is on https://htseq.readthedocs.io/en/release_0.11.1/.

Start by installing the tool.

bash

pip3 install HTSeq

Finish the following commands to calculate the counts.

bash

python3 -m HTSeq.scripts.count -m ???? -f ???? ???? ???? > 6h.counts
python3 -m HTSeq.scripts.count -m ???? -f ???? ???? ???? > 2cells.counts

You may use

bash

python3 -m HTSeq.scripts.count --help
to get help. Use union mode.

Now, inspect contents of the .counts file. Search for some of the gene names (ENSEMBL ids).

Differential expression analysis

For analysis, we will use the R library DESeq2. Load the data with.

R

library(DESeq2)
sampleFiles<-grep('counts',list.files("."),value=TRUE)
sampleCondition <- c("2cells", "6h")
sampleTable<-data.frame(sampleName=sampleFiles, fileName=sampleFiles, condition=sampleCondition)
ddsHTSeq<-DESeqDataSetFromHTSeqCount(sampleTable=sampleTable, directory=".", design=~condition)
ddsHTSeq
The last command visualized the result.

The analysis and visualization of the expression levels may be done with the following code.

R

dds <- DESeq(ddsHTSeq)
saveRDS(dds, file="dds.rds")
res <- results(dds)
res

plotMA(res, ylim=c(-10,10))

The textual input shows the difference in expression for all genes sorted by gene name. The plot shows the same result. Note the red triangle on the top. As the dataset is small, there is only a single gene that shows statistically significant differential expression. In the library vignette, you may find much more representative examples based on a study of multiple samples.

If we had multiple samples, PCA would show the clustering by test condition. Here, only a single dimension is enough. Similarly, we can visualize expressions by the condition for a gene of interest.

References

courses/bin/tutorials/tutorial9.txt · Last modified: 2020/04/20 17:56 by rysavpe1