Quality of Mapping Exercise

From wiki
Jump to: navigation, search

Introduction

We have aleady looked at read quality, but our alignment now is a new entity, and also needs to be checked for quality.

Aims

Concerns controlling mapping quality

In this part you will learn to:

  • mark duplicated read pairs
  • index BAM files
  • check the fragment size distribution
  • check the read coverage over the gene body
  • check if we have enough data

We will use the following software tools:

Loading the software

  • RSeQC is a python program and python is installed by default, so it needs no loading.
  • The other three are loaded as follows:
module load samtools BamQC picard-tools

The data set you'll be using is downloaded from ENA (http://www.ebi.ac.uk/ena/data/view/SRP019027). The reads belong to sample SRR769316. The data set is tailored with respect to the time allocated for the exercise. Reads were aligned to the first 20 Mb of chromosome 19 of the mouse reference genome (GRCm38/mm10) using TopHat.

You will use the following files:

  • m10_chr19-1-20000000_Ensembl.bed: Ensembl mouse gene models
  • SRR769316.bam, we are not using your own tophat alignment from the the previous exercise, but this one.

Data

The data is available in the directory 03_Quality_of_Mapping:

cd ~/i2rda_data/03_Quality_of_Mapping

We should see the file SRR769316.bam in there, a binary alignment file. We want to have a quick look at it using less, but bam is a binary file. It will throw gibberish on the screen, and may even reset your character encoding, which would force you to quite the session. So, we also need samtools view to view it properly. Also, it has very long lines so we will use a special option of less, the -S option.

samtools view SRR769316.bam | less -S

where:

  • | is the usual linux pipe operator
  • -S: chop long lines instead of folding them

You can navigate through a line by using the [Right arrow] and [Left arrow] keys. You can navigate through the file line-by-line using the Up-arrow and Down-arrow keys, or advance page-by-page using the Enter key, [Space-bar and B-key. You can exit the file using the Q-key.

Alignment Quality

One concern is to quickly see how many of the reads have aligned. We've talked about the SAM format's flag value, well 4 is the value we want to check.

samtools view -F 4 -c SRR769316.bam

Unmapped reads are given by

samtools view -f 4 -c SRR769316.bam

Simon Andrews' lab gave us FastQC which we saw in the beginning, and BamQC is also from them and has a very similar way of working:

bamqc SRR769316.bam

Similarly, we need firefox to view the results whose page styling is very similar to FASTQC, although the contents are all about a different object: an alignment. This time however, we give no background explanation as before, but much can be gained by scrolling through.

firefox SRR769316_bamqc.html

As you can see this is quite a well behaved dataset and has aligned very well.

Marking duplicates

Picard is a software suite programmed in Java. It started out as a "samtools for Java", so it is very similar, but it has evolved quite alot. Its MarkDuplicates subcommand is very widely used. It marks all duplicate read pairs in our dataset. We can find out how it may be used with

java -jar $PICARDJARPATH/picard.jar MarkDuplicates I=SRR769316.bam O=SRR769316_duplicates_marked.bam M=SRR769316_duplicates.metrics.csv

where:

  • I=: input SAM or BAM file to analyze
  • O=: the output file to write marked records to
  • M=: file to write duplication metrics to

This manner of presenting inputs and outputs to a program is quite particular to picard-tools, and we also see in its widely used sister program GATK. MarkDuplicates creates two files, i.e. a bam file with all duplicate records flagged, and a file containing the duplication metrics.

  • What is the duplicate rate of sample SRR769316?

To get the result quickly, try:

awk 'BEGIN{FS="\t"}{if(NF>9) printf "%20s\t%20s\t%20s\n",$3,$7,$9}' SRR769316_duplicates.metrics.csv

Checking the fragment size distribution

RSeQC's inner_distance.py script calculates the inner distance (or insert size) between two paired RNA reads:

inner_distance.py -i SRR769316.bam -r Reference_files/mm10_chr19-1-20000000_Ensembl.bed -o SRR769316

where:

  • -i: alignment file in BAM or SAM format
  • -r: reference gene model in BED format
  • -o: prefix of output file(s)

The bed-file with reference gene models is used to correct for any introns that are present between the reads.

The script first determines the genomic DNA size (D_size) between two paired reads: D_size = read2_start - read1_end, then

  • if two paired reads map to the same exon: inner distance = D_size
  • if two paired reads map to different exons: inner distance = D_size - intron_size
  • if two paired reads map to non-exonic region (such as intron and intergenic region): inner distance = D_size

The inner distance might be a negative value if the two reads overlapped.

Open the plot:

xpdf SRR769316.inner_distance_plot.pdf &

Question:

  • Knowing that the read length is 99 bases what is the mean length of the fragments?

Checking the read coverage over the gene body

RSeQC's geneBody_coverage.py script reports the read coverage of all genes in the percentile of their length. For its intensive search however, it would like us to generate an index first:

samtools index SRR769316.bam

This will automatically create a SRR769316.bam.bai file

Now we ca go ahead an launch the script.

geneBody_coverage.py -i SRR769316.bam -r Reference_files/mm10_chr19-1-20000000_Ensembl.bed -o SRR769316

When it's finished, open the plot:

xpdf SRR769316.geneBodyCoverage.curves.pdf &

Questions:

  • Is there a 3' or 5' bias?
  • How would you explain the bell shaped curve?

Checking if we have enough data

RSeQC's junction_annotation.py script compares detected splice junctions to reference gene models:

junction_annotation.py -i SRR769316.bam -r Reference_files/mm10_chr19-1-20000000_Ensembl.bed -o SRR769316

Splicing annotation is performed at two levels: splice event level and splice junction level. All detected junctions can be grouped into 3 exclusive categories:

  1. annotated: Known junction. Both 5’ and 3' splice site are present in the reference gene model.
  2. complete_novel: Complete new junction. Neither of the two splice sites is present in the reference gene model.
  3. partial_novel: Partially new junction. Only one of the splice sites is present in the reference gene model.

Now open the plots, we use a heavier pdf viewer now to open two plots at the same time.

evince SRR769316.splice_*.pdf &

Question:

  • Is the data mostly supporting known or novel junctions?

RSeQC's junction_saturation.py script checks for saturation by resampling 5%, 10%, 15%, ..., 95% of total alignments, and then detects splice junctions from each subset and compares them to reference gene model:

junction_saturation.py -i SRR769316.bam -r Reference_files/mm10_chr19-1-20000000_Ensembl.bed -o SRR769316 -v 5

where:

  • -v: minimum number of supporting reads to call a junction (default=1)

This shows if more sequencing will enable the discovery of new events.

Open the plot:

xpdf SRR769316.junctionSaturation_plot.pdf &

Question:

  • Do you think more sequencing is needed?

Some issues are only detectable in the context of the genome:

  • Duplicate reads
  • Fragment size distribution
  • Gene coverage
  • Completeness of data