Single Nucleotide Variant (SNV) calling Tutorial GVA2023


Overview:

SAMtools is a suite of commands for dealing with databases of mapped reads. You'll be using it quite a bit throughout the course. It includes programs for performing variant calling (mpileup-bcftools). This tutorial expects you have already completed the Mapping tutorial.

Learning Objectives

  1. Work with a more complex conda installation, and how to troubleshoot it.
  2. Familiarize yourself with SAMtools.
  3. Use SAMtools to identify variants in the E. coli genomes we mapped in the previous tutorial.

Installing SAMtools 

As we have done with: fastqc, cutadapt, and bowtie2, we want to install samtools and bcftools into a new environment (we'll call this one GVA-SNV). Once again, having access to conda-forge will be required to install the most recent version. 

 See if you can use what you know from installing bowtie to install the new environment without the additional help contained in this drop down. The code will still be hidden if you need a hint.
  •  https://anaconda.org/bioconda/samtools and https://anaconda.org/bioconda/bcftools both show the bioconda channel as being the best source for installing the programs. 
  • The instructions above list having the second channel "conda-forge" as being required to install the most recent versions
  • In the read mapping tutorial we saw that we could install programs to a new environment when we created said environment without needing the install keyword
Complete answer for installing samtools and bcf tools
conda create -n GVA-SNV -c conda-forge -c bioconda samtools bcftools

for reasons discussed https://github.com/bioconda/bioconda-recipes/issues/34190 and https://edcarp.github.io/introduction-to-conda-for-data-scientists/03-using-packages-and-channels/index.html listing conda-forge before bioconda is critical.


 Expand here for detailed descriptions of the troubleshooting that took place for this last year.


The assumption last year was that the correct command would be: conda install -c bioconda samtools as it is what was listed at https://anaconda.org/bioconda/samtools. Instead the correct command ended up being: conda install -c bioconda samtools bcftools openssl=1.0 

There are 2 different things going on in this command.

  1. Forcing the installation of a specific version of openssl. In this case, a lower version than would normally be installed if samtools were installed by itself. According to https://github.com/bioconda/bioconda-recipes/issues/12100 my understanding is that when the conda package was put together there is an error wherein samtools specifically says to get version 1.1 of openssl, but the samtools program specifically requires version 1.0 to be present.
  2. We are installing both samtools and bcftools at the same time. This can clean up some installation problems when there are conflicts between individual packages and you want to use them in a single environment. An alternative would be to have a samtools environment and a bcftools environment, but that creates unnecessary steps of having to change environments in the middle of your analysis.
 Click here to expand and see what the outcome of the assumed installation command is, what the problem is, and steps you could take to fix it.

This box contains example commands and outputs showing you something that does NOT work for educational and diagnostic purposes. If you use the code listed in this box, be sure you use ALL the code or you may run into downstream problems with this tutorial.

conda install -c bioconda samtools
samtools --version

The above command appeared to install correctly as other conda installations did, but the second command which you would expect to show the version of samtools instead returns the following error:

samtools: error while loading shared libraries: libcrypto.so.1.0.0: cannot open shared object file: No such file or directory

Googling the entire error the top results clearly mentioned conda and several pages listed problems associated "fixes" with different conda installation commands:

  1. One of the suggested fixes was to add access to the conda-forge channel (as we have done this year).
  2. Additionally, last year the entire course was taught with the hope of sequentially adding new programs to a single growing environment. As mentioned yesterday, such an approach is not always optimal/easy, and hence why this year we are creating a number of additional environments. Working from the assumption that you wanted to keep a single environment, one fix that appeared to be working well based on community feedback was conda install -c bioconda samtools=1.9 --force-reinstall, but at the expense of altering existing packages in the environment. When running the command, the number of programs that would be downgraded was nearly 2 full screens long.
  3. Rather than jumping to the "force-reinstall" solution it was suggested to copy the existing conda environment to a new "test" environment (conda create --name GVA2021-samtools-test --clone GVA2021). Once in the new environment, using the "force-reinstall" command above would have given access to samtools, but would then also require testing other programs in the environment (such as bowtie2, cutadapt, fastqc). Assuming all programs still (seemed) to work you could then rename the environment (conda create --name GVA2021-V2 --clone GVA2021-samtools-test; conda env remove --name GVA2021-samtools-test). Obviously the more programs that are added, the more likely running into these kinds of conflicts (where different versions of the same dependency are required), and the more programs you would have to check to see if the "test" envvironment broke any previously installed programs. 
  4. As there was information that suggested the issues were specific to version 1.12 (including: https://github.com/bioconda/bioconda-recipes/issues/13958), another solution was simply to install an older version of samtools deliberately from the start. In order to do this, I first had to remove the existing (incorrect) samtools version (conda remove samtools) and then specify the older version we wanted to use (conda install -c bioconda samtools==1.11). This then allowed the samtools --version command to give expected output of version 1.11 Unfortunately, during testing it quickly became obvious that the next program to install (bcftools) was going to create an entirely new set of installation problems meaning that samtools again hat to be uninstalled and samtools bcftools and the downgraded openssl version installed together. conda remove samtools ; conda install -c bioconda samtools bcftools openssl=1.0. 


Make sure you have the expected versions of samtools and bcftools installed
samtools --version
bcftools --version


samtools --version output:

samtools 1.17
Using htslib 1.17
Copyright (C) 2023 Genome Research Ltd.
# Followed by a bunch of compilation details.


bcftools --version output:

bcftools 1.17
Using htslib 1.17
Copyright (C) 2023 Genome Research Ltd.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.

If you get something else, such as an error message with bcftools --version:

"bcftools: error while loading shared libraries: libgsl.so.25: cannot open shared object file: No such file or directory" 

do not proceed, check your conda install command and if it matches the one listed above, get my attention.

Calling variants in reads mapped by bowtie2

Prepare your directories

Since the $SCRATCH directory on stampede2 is effectively infinite for our purposes, we're going to copy the relevant files from our mapping tutorial into a new directory for this tutorial. This should help you identify what files came from what tutorial if you look back at it in the future. Let's copy over just the read alignment file in the SAM format and the reference genome in FASTA format to a new directory called GVA_samtools_tutorial.

Check your work or get the answer here
cds
mkdir GVA_samtools_tutorial
cd GVA_samtools_tutorial
cp $SCRATCH/GVA_bowtie2_mapping/bowtie2/SRR030257.sam .
cp $SCRATCH/GVA_bowtie2_mapping/NC_012967.1.fasta .

Unexpected output when you try to copy final files from mapping tutorial

If you see messages saying something similar to the following:

cp: cannot stat '/scratch/01821/ded/GVA_bowtie2_mapping/bowtie2/SRR030257.sam': No such file or directory
cp: cannot stat '/scratch/01821/ded/GVA_bowtie2_mapping/NC_012967.1.fasta': No such file or directory

It suggests something you either did not yet complete the mapping tutorial, or more likely, you stored these files in a different directory. If you think you completed the mapping tutorial, get my attention and be ready to share your screen and I'll try to help you find your missing files.

When copy commands execute successfully, the expected output is silent (no output at all)

Index the FASTA reference file

Assuming you have the above output for samtools --version and bcftools --version (both 1.17), first, you need to index the reference file. (This isn't the same as indexing it for read mapping. It's indexing it so that SAMtools can quickly jump to a certain base in the reference.)

Command to index the reference file for SAMtools
samtools faidx NC_012967.1.fasta

Take a look at the new *.fai file that was created by this command see if you have any idea what some of the numbers mean. 

Alternative to head/tail/cat for examining a file without causing programs to crash
less NC_012967.1.fasta.fai  # can exit with "q"

As you can see, the less command also works perfectly well with files that are not in danger of crashing anything without cluttering your terminal  with lines of a file.

Convert mapped reads from SAM to BAM, sort, and index

SAM is a text file, so it is slow to access information about how any given read was mapped. SAMtools and many of the commands that we will run later work on BAM files (essentially GZIP compressed binary forms of the text SAM files). These can be loaded much more quickly. Typically, they also need to be sorted, so that when the program wants to look at all reads overlapping position 4,129,888, it can easily find them all at once without having to search through the entire BAM file.

The following 3 commands are used to:

  1. convert from SAM to BAM format
  2. sort the BAM file
  3. index the sorted BAM file

As you might guess this is computationally intense and as such must be iDEV node or submitted as a job (more on this on Friday). If you want to submit them to the job queue, you will want to separate them with a ";" to ensure that they run sequentially rather than simultaneously as each uses the output of the previous command. Under no circumstances should you run this on the head node.

Do not run on head node

Use hostname to verify you are still on the idev node. expect to see a computer number (NOT a login number) in front of 

stampede2.tacc.utexas.edu

If not, and you need help getting a new idev node, see this tutorial.

Commands to be executed in order...
samtools view --threads 48 -b -S -o SRR030257.bam SRR030257.sam
samtools sort --threads 48 SRR030257.bam -o SRR030257.sorted.bam
samtools index -@ 48 SRR030257.sorted.bam
 view sort and index are all subcommands to the samtools program, and --help can be invoked to get more information about each of the subcommands. Can you figure out what the purpose of each of the above options is?
subcommandflagvaluepurpose
view/sort--threads48use 48 additional threads
view-bhas no valueis a toggle to output in bam format
view-Shas no valuewas a toggle to declare the input format as sam. help now tells you that this is depreciated as the program auto detects input format
view-oSRR030257.bamwrite the output of the command to the SRR030257.bam file
view
SRR030257.samunflagged option/keyword. in this case, the top line of the help output lists:
Usage: samtools view [options] <in.bam>|<in.sam>|<in.cram> [region ...]

the <>|<>|<> section states that you can give a bam, sam, or cram file as input

sort
SRR030257.bamunflagged option/keyword. in this case, the top line of the help output lists:

Usage: samtools sort [options...] [in.bam]

the in.bam section states that the input must be in bam format.
Note that the input to the sort subcommand is the output of the view command.

sort-oSRR030257.sorted.bamwrite output to file SRR030257.sorted.bam
index-@48use 48 additional threads. note that in both the previous help outputs, they listed -@, --threads because both are recognized as the same flag. in the first 2 cases, we used the --threads but could have used -@ but for the index subcommand, only -@ was allowed.
index
SRR030257.sorted.bam

unflagged option/keyword. in this case, the top line of the help output lists:
Usage: samtools index [-bc] [-m INT] <in.bam> [out.index]

in this case we are again using the previous commands output file as an input file. 
Note that we have not listed an output file here. Keep that in mind as you read further on.


Why was they -S option included in the view command above if its use is depreciated?

This is included to highlight 2 things:

  1. sometimes you find or are given commands that work, but you dont necessarily understand why they work.
  2. sometimes program version updates change options in ways that dont require you to update your commands

Cheat Sheat

This "view" "sort" "index" series is a really common sequence of commands, so you might want to add it to your personal cheat sheet if you are keeping one.

It is expected that the first command generate no output, the second command to generate a single line from the bam_sort_core regarding files and memory blocks, and the third line to again generate no output. While the first 2 commands take a few minutes each and the third command is very quick on a single thread, with 48 additional threads all complete very quickly.

Examine the output of the previous commands to get an idea of whats going on. Here are some prompts of how to do that:

List the contents of the output directory to see what new files have been created
ls -1  # This is the number 1 not the letter L. Several people seem to get confused about this each year, it is not necessary for any reason other than to give a single column of output regardless of window size.
Expected output
NC_012967.1.fasta
NC_012967.1.fasta.fai
SRR030257.bam
SRR030257.sam
SRR030257.sorted.bam
SRR030257.sorted.bam.bai
 Given what we saw above of fasta index file gaining a .fai ending, can you guess what a *.bai file is?

Sure enough, it's the index file for the BAM file. Since we did not specify and output file in our index command, it was assumed we wanted to simply append ".bai" corresponding to its new type of a BAI-format index of our input bam file.

You might be tempted to gzip BAM files when copying them from one computer to another. Don't bother! They are already internally compressed, so you won't be able to shrink the file. Further, to the best of my knowledge, no programs accept a gzipped bam file as a format to use.

On the other hand, compressing SAM files will save some space, but the conversion between bam back to sam is pretty simple/quick. Making storage of bam files likely a better decision.

Call genome variants

Now we use the mpileup command from samtools to compile information about the bases mapped to each reference position. The output is a BCF file. This is a binary form of the text Variant Call Format (VCF). For more information about VCF files: https://docs.gdc.cancer.gov/Data/File_Formats/VCF_Format/

You should execute this command as is without trying to determine the options for yourself as even with 64 threads it will take several minutes to complete. During the completion time you can review the different options used.
bcftools mpileup
bcftools mpileup --threads 48 -O u -f NC_012967.1.fasta SRR030257.sorted.bam -o SRR030257.bcf


 Using the information from the bcftools mpileup command without any options, can you figure out what are all the options doing?
Optionpurpose

--threads 48

use 48 additional threads
-f NC_012967.1.fasta

reference sequence file that has a corresponding faidx index .fai file

SRR030257.sorted.bamBAM input file to calculate pileups from
-O ugenerates uncompressed BCF output
-o SRR030257.bcfOutput file SRR030257.bcf

Historical command (now depreciated) which used to give nearly the same output as the bcftools mpileup command

Last year, in addition to the bcftools mpileup command listed above, a second command a second command using samtools mpileup was listed as an option. This was a very common command structure that you may come across elsewhere  (samtools mpileup -u -f NC_012967.1.fasta SRR030257.sorted.bam > SRR030257.bcf). note that besides using the program samtools instead of bcftools, the only differences are the use of -u instead of -O u, and piping the output (>) to the SRR30257.bcf file instead of naming the file with a -o flag.

Last year it was noted that if you tried the samtools command, there was a  warning stating that:

"samtools mpileup option `u` is functional, but deprecated. Please switch to using bcftools mpileup in future."

Further, I warned that adjusting to the new command would have value as typically once programers start warning that functionality is "depreciated" it is only a matter of time before it is "no longer supported" and then just flat out "broken". Sure enough, in less than a year's time, it is no longer working. This is one of the reasons why updating to the newest version of a program is not always recommended if the version you are using is working for you (more on Friday).


Sending programs to the background.

The samtools mpileup command will take a few minutes to run even with 48 threads. If you have read through the information about the different options, as practice for a fairly common occurrence when working with the iDEV environment, you could try putting it in the background by pressing control-z and then typing the command bg so that you can do some other things in this terminal window at the same time. Remember, there are still many other processors available on this node for you to do other things! Just remember that if you have something running in the background you need to check in with it to see if it is still running with the ps command or watch the command line every time you execute a new command as you may see information about your background task having finished.

The fg command (foreground) is the opposite of the bg (background) command. If you want to return your command to your active prompt so you are notified directly when the command finishes (or errors) simply type 'fg' assuming you only have 1 job running in the background.


Convert genome variants to human readable format

Once the mpileup command is complete, convert the BCF file to a "human-readable" VCF file using a bcftools command.

bcftools call -v -c SRR030257.bcf > SRR030257.vcf

What are these options doing?

 See if you can start with the base command "bcftools" and figure out what each option above is doing on the bcftools command we used above. Click here when you think you know.
Optionpurpose
callspecific subcommand to be executed by bcftools for calling SNP and indels
-v

output potential variant sites only

-cconsensus calling
SRR030257.bcf
input bcf file
> SRR030257.vcf
output as a vcf file

If you are especially observant, you might notice that in the bcftools mpileup options, the output type option (-O) had an option v which lists that the output file would have been generated as a uncompressed vcf file. While this may seem like you could have used -O v instead of -O u and skipped a step, note that the mpileup subcommand lacked options for only outputting variant sties or using consensus calling.

Take a look at the SRR030257.vcf file using less. It has a nice header explaining what the columns mean, including answers to some of your questions from yesterday's presentations. https://docs.gdc.cancer.gov/Data/File_Formats/VCF_Format/ can be used to figure out the columns are and what types of information they provide. Below this are the rows of data describing potential genetic variants.

Analyzing variants detected

VCF format has alternative Allele Frequency tags denoted by AF= Try the following command to see what frequency our variants exist at.

grep AF1 SRR030257.vcf

If you look at the AF1= values you will see all the lines are either ~ 0.5, or 1.

You can see that even easier with this handy awk 1 liner
 awk -F";" '{for(i=1;i<=NF;i++){if ($i ~ /AF1/){print $i}}}' SRR030257.vcf

^splits each line into columns based on where the ";"s, then searches through each column, if the "AF1" is found in the column, that column is printed. From the output it is even clearer that frequencies are coming up.

Cheat Sheat

This is a pretty useful awk 1 liner to keep track of for pulling a single column out when you know it will have an identifier but not what column it will occur in; just substitute what you expect to find for "AF1" and substitiute whatever character you want to use to break the line up for -F";" (with -F"\t" for tab or -F"," being the 2 most common options for tsv and csv files respectively).
The above result can be piped into sort to sort the lines into increasing order
 awk -F";" '{for(i=1;i<=NF;i++){if ($i ~ /AF1/){print $i}}}' SRR030257.vcf | sort
 Did you see anything when you were running the commands in this tutorial to suggest that it was looking for diploid genomes?

After you ran the bcftools call command you saw: "Note: none of --samples-file, --ploidy or --ploidy-file given, assuming all sites are diploid". Just like on a webpage you can use control/command + F to find specific text in the window. Look for 'diploid' and you should see the line referenced above.

Obviously this suggests a way that you could go back and reanalyze this data introducing one of the recommended flags to the bcftools call command and see how this might effect your analysis. If you choose to do this, I suggest adding descriptive file name between 'SRR030257' and '.vcf' to make the results easier to compare.

 Rather than (or in addition to) going back and changing the analysis to look for variation from our haploid genome, can you come up with a way to create a filtered .vcf file that only has variants with a frequency of 1?

An initial attempt might be something like this:

Note these 2 examples will give the exact same output. The first command gives you extra practice with piping, and may help order your thoughts somewhat.
cat SRR030257.vcf | grep AF1=1 > SRR030257.filtered.vcf


grep AF1=1 SRR030257.vcf > SRR030257.filtered.vcf

This is does give us exactly what we asked for: all the lines that show a variant allele frequency of 1. Unfortunately, we lost all the useful header information at the top of the original  SRR030257.vcf file. 

 Can you come up with a way to include this information possibly by looking at the help information on grep?

From 'grep -h' you can see the that the '-v' inverts the match so it will print everything EXCEPT for lines that match.

2 equivalent answers again
cat SRR030257.vcf | grep -v AF1=0 > SRR030257.filtered.vcf

grep -v AF1=0 SRR030257.vcf > SRR030257.filtered.vcf

Will preserve all lines that don't have 'AF1=0' value on the line and is one way of doing this. If you look closely at the non-filtered file you will see that the frequencies are given as AF1=0.### so by filtering out lines that have 'AF1=0' in them we get rid of all frequencies that are not 1, including say 'AF1=0.99'. 

 How you would change this to variants that have a frequency of at least 90%?

Remember from our Raw Sequencing Data tutorial yesterday that we can group certain characters together by placing them between square brackets [].

2 equivalent answers again
cat SRR030257.vcf | grep -v AF1=0.[0-8] > SRR030257.filtered.vcf

grep -v AF1=0.[0-8] SRR030257.vcf > SRR030257.filtered.vcf

Here we added a decimal point after the 0, and then allowed for a match to any digit between 0 and 8. Thus lines that have AF1=1 would not match, nor would a line with AF1=0.9 . You might make a note to think back on this after tomorrow's presentation covering when we should believe variants are real.


Return to GVA2023 course page.


Optional Exercises at the end of class or for Wednesday/Thursday choose your own tutorial time.

Calling variants in trimmed reads.

  1. Trim both Read1 and Read2 using info from read preprocessing tutorial.
  2. Map reads with bowtie2 using info from read mapping tutorial.
  3. Call variants using this tutorial.

Remember in the intro tutorial we talked about file/directory naming. Be sure you don't write over your old files. Maybe create a new directories like GVA_samtools_bowtie_improved for the outputs.

Further Optional Exercises

  • If you use additional mapping programs, which mapper finds more variants?
  • Can you figure out how to filter the VCF files on various criteria, like coverage, quality, ... ?
  • How many high quality mutations are there in these E. coli samples relative to the reference genome?
  • Look at how the reads supporting these variants were aligned to the reference genome in the Integrative Genomics Viewer (IGV). This will be a separate tutorial for tomorrow.


Return to GVA2023 course page.