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.
- Work with a more complex conda installation, and how to troubleshoot it.
- Familiarize yourself with SAMtools.
- Use SAMtools to identify variants in the E. coli genomes we mapped in the previous tutorial.
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.
- 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
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.
- 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.
- 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.
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.
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:
Googling the entire error the top results clearly mentioned conda and several pages listed problems associated "fixes" with different conda installation commands:
- One of the suggested fixes was to add access to the conda-forge channel (as we have done this year).
- 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.
- 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.
- 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.
samtools --version output:
bcftools --version output:
Calling variants in reads mapped by bowtie2
Prepare your directories
Since the $SCRATCH directory on lonestar 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.
Unexpected output when you try to copy final files from mapping tutorial
If you see messages saying something similar to the following:
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 1.15.1), 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.)
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.
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:
- convert from SAM to BAM format
- sort the BAM file
- 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
If not, and you need help getting a new idev node, see this tutorial.
|view/sort||--threads||64||use 64 additional threads|
|view||-b||has no value||is a toggle to output in bam format|
|view||-S||has no value||was 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||-o||SRR030257.bam||write the output of the command to the SRR030257.bam file|
|view||SRR030257.sam||unflagged 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.bam||unflagged 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.
|sort||-o||SRR030257.sorted.bam||write output to file SRR030257.sorted.bam|
|index||-@||64||use 64 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.|
unflagged option/keyword. in this case, the top line of the help output lists:
in this case we are again using the previous commands output file as an input file.
Why was they -S option included in the view command above if its use is depreciated?
This is included to highlight 2 things:
- sometimes you find or are given commands that work, but you dont necessarily understand why they work.
- sometimes program version updates change options in ways that dont require you to update your commands
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 64 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:
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/
|use 64 additional threads|
reference sequence file that has a corresponding faidx index .fai file
|SRR030257.sorted.bam||BAM input file to calculate pileups from|
|-O u||generates uncompressed BCF output|
|-o SRR030257.bcf||Output 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
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, and in less than 4 version updates, 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.
samtools mpileup command will take a few minutes to run even with 64 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.
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.
What are these options doing?
|call||specific subcommand to be executed by bcftools for calling SNP and indels|
output potential variant sites only
|input bcf file|
|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.
If you look at the AF1= values you will see all the lines are either ~ 0.5, or 1.
^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.
-F"\t"for tab or
-F","being the 2 most common options for tsv and csv files respectively).
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.
An initial attempt might be something like this:
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.
From 'grep -h' you can see the that the '-v' inverts the match so it will print everything EXCEPT for lines that match.
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'.
Remember from our Raw Sequencing Data tutorial yesterday that we can group certain characters together by placing them between square brackets .
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.
Optional Exercises at the end of class or for Wednesday/Thursday choose your own tutorial time.
Calling variants in trimmed reads.
- Trim both Read1 and Read2 using info from read preprocessing tutorial.
- Map reads with bowtie2 using info from read mapping tutorial.
- 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.