Short tutorial¶
Here we provide a short tutorial that guides you through the main features of Snakemake. Note that this is not suited to learn Snakemake from scratch, rather to give a first impression. To really learn Snakemake (starting from something simple, and extending towards advanced features), use the main Snakemake Tutorial.
This document shows all steps performed in the official Snakemake live demo, such that it becomes possible to follow them at your own pace. Solutions to each step can be found at the bottom of this document.
The examples presented in this tutorial come from Bioinformatics. However, Snakemake is a general-purpose workflow management system for any discipline. For an explanation of the steps you will perform here, have a look at Background. More thorough explanations are provided in the full Snakemake Tutorial.
Prerequisites¶
First, install Snakemake via Conda, as outlined in Installation via Conda/Mamba. The minimal version of Snakemake is sufficient for this demo.
Second, download and unpack the test data needed for this example from here, e.g., via
mkdir snakemake-demo
cd snakemake-demo
wget https://github.com/snakemake/snakemake-tutorial-data/archive/v5.4.5.tar.gz
tar --wildcards -xf v5.4.5.tar.gz --strip 1 "*/data"
Step 1¶
First, create an empty workflow in the current directory with:
mkdir workflow
touch workflow/Snakefile
Once a Snakefile is present, you can perform a dry run of Snakemake with:
snakemake -n
Since the Snakefile is empty, it will report that nothing has to be done. In the next steps, we will gradually fill the Snakefile with an example analysis workflow.
Step 2¶
The data folder in your working directory looks as follows:
data
├── genome.fa
├── genome.fa.amb
├── genome.fa.ann
├── genome.fa.bwt
├── genome.fa.fai
├── genome.fa.pac
├── genome.fa.sa
└── samples
├── A.fastq
├── B.fastq
└── C.fastq
You will create a workflow that maps the sequencing samples in the
data/samples
folder to the reference genome data/genome.fa
.
Then, you will call genomic variants over the mapped samples, and create
an example plot.
First, create a rule called map_reads
, with input files
data/genome.fa
data/samples/A.fastq
and output file
results/mapped/A.bam
To generate output from input, use the shell command
"bwa mem {input} | samtools view -Sb - > {output}"
Providing a shell command is not enough to run your workflow on an
unprepared system. For reproducibility, you also have to provide the
required software stack and define the desired version. This can be done
with the Conda package manager, which is directly
integrated with Snakemake: add a directive
conda: "envs/mapping.yaml"
that points to a Conda environment
definition,
with the following content
channels:
- bioconda
- conda-forge
dependencies:
- bwa =0.7.17
- samtools =1.9
Upon execution, Snakemake will automatically create that environment, and execute the shell command within.
Now, test your workflow by simulating the creation of the file
results/mapped/A.bam
via
snakemake --use-conda -n results/mapped/A.bam
to perform a dry-run and
snakemake --use-conda results/mapped/A.bam --cores 1
to perform the actual execution.
Step 3¶
Now, generalize the rule map_reads
by replacing the concrete sample name
A
with a wildcard {sample}
in input and output file the rule
map_reads
. This way, Snakemake can apply the rule to map any of the three
available samples to the reference genome.
Test this by creating the file results/mapped/B.bam
.
Step 4¶
Next, create a rule sort_alignments
that sorts the obtained .bam
file by
genomic coordinate. The rule should have the input file
results/mapped/{sample}.bam
and the output file
results/mapped/{sample}.sorted.bam
and uses the shell command
samtools sort -o {output} {input}
to perform the sorting. Moreover, use the same conda:
directive as
for the previous rule.
Test your workflow with
snakemake --use-conda -n results/mapped/A.sorted.bam
and
snakemake --use-conda results/mapped/A.sorted.bam --cores 1
Step 5¶
Now, we aggregate over all samples to perform a joint calling of genomic variants. First, we define a variable
samples = ["A", "B", "C"]
at the top of the Snakefile
. This serves as a definition of the
samples over which we would want to aggregate. In real life, you would
want to use an external sample sheet or a config
file
for things like this.
For aggregation over many files, Snakemake provides the helper function
expand
(see the
docs).
Create a rule call
with input files
fa="data/genome.fa"
bam=expand("results/mapped/{sample}.sorted.bam", sample=samples)
output file
"results/calls/all.vcf"
and shell command
bcftools mpileup -f {input.fa} {input.bam} | bcftools call -mv - > {output}
Further, define a new conda environment file with the following content:
channels:
- bioconda
- conda-forge
dependencies:
- bcftools =1.9
Step 6¶
Finally, we strive to calculate some exemplary statistics. This time, we don’t use a shell command, but rather employ Snakemake’s ability to integrate with scripting languages like R and Python, and Jupyter notebooks.
First, we create a rule plot_quals
with input file
"results/calls/all.vcf"
and output file
"results/plots/quals.svg"
.
Instead of a shell command, we use Snakemake’s Jupyter notebook integration by specifying
notebook:
"notebooks/plot-quals.py"
instead of using the shell
directive as before.
Next, we have to define a conda environment for the rule, say
workflow/envs/stats.yaml
, that provides the required Python packages to
execute the script:
channels:
- bioconda
- conda-forge
dependencies:
- pysam =0.17
- altair =4.1
- altair_saver =0.5
- pandas =1.3
- jupyter =1.0
Then, we let Snakemake generate a skeleton notebook for us with
snakemake --draft-notebook results/plots/quals.svg --cores 1 --use-conda
Snakemake will print instructions on how to open, edit and execute the notebook.
We open the notebook in the editor and add the following content
import pandas as pd
import altair as alt
from pysam import VariantFile
quals = pd.DataFrame({"qual": [record.qual for record in VariantFile(snakemake.input[0])]})
chart = alt.Chart(quals).mark_bar().encode(
alt.X("qual", bin=True),
alt.Y("count()")
)
chart.save(snakemake.output[0])
As you can see, instead of writing a command line parser for passing
parameters like input and output files, you have direct access to the
properties of the rule via a magic snakemake
object, that Snakemake
automatically inserts into the notebook before executing the rule.
Make sure to test your workflow with
snakemake --use-conda --force results/plots/quals.svg --cores 1
Here, the force ensures that the readily drafted notebook is re-executed even if you had already generated the output plot in the interactive mode.
Step 7¶
So far, we have always specified a target file at the command line when
invoking Snakemake. When no target file is specified, Snakemake tries to
execute the first rule in the Snakefile
. We can use this property to
define default target files.
At the top of your Snakefile
define a rule all
, with input files
"results/calls/all.vcf"
"results/plots/quals.svg"
and neither a shell command nor output files. This rule simply serves as an indicator of what shall be collected as results.
Step 8¶
As a last step, we strive to annotate our workflow with some additional information.
Automatic reports¶
Snakemake can automatically create HTML reports with
snakemake --report report.html
Such a report contains runtime statistics, a visualization of the workflow topology, used software and data provenance information.
In addition, you can mark any output file generated in your workflow for
inclusion into the report. It will be encoded directly into the report,
such that it can be, e.g., emailed as a self-contained document. The
reader (e.g., a collaborator of yours) can at any time download the
enclosed results from the report for further use, e.g., in a manuscript
you write together. In this example, please mark the output file
"results/plots/quals.svg"
for inclusion by replacing it with
report("results/plots/quals.svg", caption="report/calling.rst")
and adding a
file report/calling.rst
, containing some description of the output
file. This description will be presented as caption in the resulting
report.
Threads¶
The first rule map_reads
can in theory use multiple threads. You can make
Snakemake aware of this, such that the information can be used for
scheduling. Add a directive threads: 8
to the rule and alter the
shell command to
bwa mem -t {threads} {input} | samtools view -Sb - > {output}
This passes the threads defined in the rule as a command line argument
to the bwa
process.
Temporary files¶
The output of the map_reads
rule becomes superfluous once the sorted
version of the .bam
file is generated by the rule sort
.
Snakemake can automatically delete the superfluous output once it is not
needed anymore. For this, mark the output as temporary by replacing
"results/mapped/{sample}.bam"
in the rule bwa
with
temp("results/mapped/{sample}.bam")
.
Solutions¶
Only read this if you have a problem with one of the steps.
Step 2¶
The rule should look like this:
rule map_reads:
input:
"data/genome.fa",
"data/samples/A.fastq"
output:
"results/mapped/A.bam"
conda:
"envs/mapping.yaml"
shell:
"bwa mem {input} | samtools view -b - > {output}"
Step 3¶
The rule should look like this:
rule map_reads:
input:
"data/genome.fa",
"data/samples/{sample}.fastq"
output:
"results/mapped/{sample}.bam"
conda:
"envs/mapping.yaml"
shell:
"bwa mem {input} | samtools view -b - > {output}"
Step 4¶
The rule should look like this:
rule sort_alignments:
input:
"results/mapped/{sample}.bam"
output:
"results/mapped/{sample}.sorted.bam"
conda:
"envs/mapping.yaml"
shell:
"samtools sort -o {output} {input}"
Step 5¶
The rule should look like this:
samples = ["A", "B", "C"]
rule call_variants:
input:
fa="data/genome.fa",
bam=expand("results/mapped/{sample}.sorted.bam", sample=SAMPLES)
output:
"results/calls/all.vcf"
conda:
"envs/calling.yaml"
shell:
"bcftools mpileup -f {input.fa} {input.bam} | bcftools call -mv - > {output}"
Step 6¶
The rule should look like this:
rule plot_quals:
input:
"results/calls/all.vcf"
output:
"results/plots/quals.svg"
conda:
"envs/stats.yaml"
notebook:
"notebooks/plot-quals.py.ipynb"
Step 7¶
The rule should look like this:
rule all:
input:
"results/calls/all.vcf",
"results/plots/quals.svg"
It has to appear as first rule in the Snakefile
.
Step 8¶
The complete workflow should look like this:
SAMPLES = ["A", "B", "C"]
rule all:
input:
"results/calls/all.vcf",
"results/plots/quals.svg"
rule map_reads:
input:
"data/genome.fa",
"data/samples/{sample}.fastq"
output:
"results/mapped/{sample}.bam"
conda:
"envs/mapping.yaml"
shell:
"bwa mem {input} | samtools view -b - > {output}"
rule sort_alignments:
input:
"results/mapped/{sample}.bam"
output:
"results/mapped/{sample}.sorted.bam"
conda:
"envs/mapping.yaml"
shell:
"samtools sort -o {output} {input}"
rule call_variants:
input:
fa="data/genome.fa",
bam=expand("results/mapped/{sample}.sorted.bam", sample=SAMPLES)
output:
"results/calls/all.vcf"
conda:
"envs/calling.yaml"
shell:
"bcftools mpileup -f {input.fa} {input.bam} | bcftools call -mv - > {output}"
rule plot_quals:
input:
"results/calls/all.vcf"
output:
"results/plots/quals.svg"
conda:
"envs/stats.yaml"
notebook:
"notebooks/plot-quals.py.ipynb"