VariantFrequency

Search for variant in ENA/SRA to determine allele frequency

This is an automated Snakemake workflow to search variant allele frequency in a given genome.

At a minimum, it requires Snakemake, Python3 and Pandas. It is also recommended to have Conda installed. Otherwise, you need all software listed here

Get Started

Install Conda

If you don't already have Conda, it is highly recommended that you install it. This workflow uses Conda to manage rule environments. Although you can still run it without Conda (as long as you make sure all software listed here and their dependencies are installed).

To install Conda:

echo source ~/.bashrc >> ~/.bash_profile
curl -LO https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

Log out and log back in again to activate the base conda environment.

Install Snakemake, Python3, and Pandas into base environment:

conda install -c conda-forge -c bioconda snakemake python=3.9 pandas

Adjust configurations

The main snakemake workflow is in Snakefile inside the main directory. Each submodule is stored in rules/.

The workflow starts by downloading reference genome FASTA file. The genome can be specified in config.yaml:

genome: hg38

Make sure that the genome name is exactly same as in corresponding database (eg. equCab3, hg19, mm10, etc)

Alternatively, if you have a reference genome FASTA file already, you can also put it under data/ directory with the name genome.fa. This will bypass the step to download reference genome.

A list of variants to search is stored in data/variants.bed. Modify this list to include the variant positions you want to search. Keep in mind that bed files are 0-based (each chromosome positions starts at 0, the first position on chr1 should be chr1:0-1). This file should have three columns, tab-delimited, with no header.

A list of accessions is stored in data/accessions.tsv. This is a tab-delimited text file with at least the following columns:

- sample_accession
- run_accession

run_accession allows us to download fastq files for each run while sample_accession allows us to combine different runs of same sample together. ENA_curl.sh contains an example ENA query you can use. Modify the query string to match your criteria. To generate an accession file, simply run: ./ENA_curl.sh > data/accessions.tsv

Running the workflow

To start the Snakemake workflow on a local machine or a dedicated server with Conda:

snakemake --cores n_cores --use-conda &> log.txt

Without Conda:

snakemake --cores n_cores &> log.txt

n_cores is the number of threads Snakemake is allowed to use for this workflow.

After the workflow successfully completes, an output file output.vcf can be found in Results/VCF/ #Update this after adding rule to calculate AF

Running on HPC

You can also run this workflow on an HPC cluster. Read here for more details. Most rules in this workflow have some common resources defined (memory, cpus, and time) under resources directive. You can access them using these keywords:

mem_mb: memory in MB
cpus: threads
time_min: time in minutes

Additionally, a partition keyword is defined in params directive. Since this is usually cluster-specific, you will likely want to modify the getPartition function defined in the main Snakefile. The current configuration should work out-of-box on UC Davis Farm cluster.

Performance

1. It's taking too long
   This workflow downloads raw fastq files available on SRA/ENA database and aligns them to a list of reference sequences containing 1kb up- and down-stream of each variant site. The main bottle neck is at the download step. If storage space is not a concern, it is advisable to keep these fastq files so future runs can skip the download step. (The default behaviour is to remove them after alignment to save space.) To do this: delete the temp() directive surrounding the below lines in rules/prepareFiles.smk (line 90-91)

        r1 = temp(workDir + "/Results/temp_fastq/{sample}_R1.fq"),
        r2 = temp(workDir + "/Results/temp_fastq/{sample}_R2.fq")

2. The Snakemake hangs for a long time with no output If the accession list is too long, Snakemake can take a long time to build DAG for the workflow. This is mostly due to ecessive queries on the drive for output files. You can improve performance by running the workflow in batches:

snakemake --cores n_cores --use-conda --batch genotypeGVCF=1/3

will separate the workflow into 3 batches and run the first batch. Once that is completed, you can then run:

snakemake --cores n_cores --use-conda --batch genotypeGVCF=2/3

and finally:

snakemake --cores n_cores --use-conda --batch genotypeGVCF=3/3

Troubleshooting

1. The workflow failed Snakemake will report when a job fails. Look for error message in the snakemake output (log.txt). There are different typical suspects when some of the rules fail. See more details below.
2. The rule getFASTQ fails: This is usually because the connection to ENA database is unstable or interrupted. This can be seen as the job will report errors along the lines of "failed to download xxx" Reruning the workflow will usually fix this issue. Some other reasons for this issue could be that certain accession doesn't have paired-end fastq files available. This should also be reported in the error message. It is recommended to then remove those accessions from accessions.tsv file.