Pipeline Container
In order to make the pipeline as portable as possible across computing environments an Apptainer container is supplied. This section gives a brief description on how to use the container to analyse your data, and you should refer to the corresponding non-container sections for more details.
All of the scripts below will need to be amended for use with the HPC machine you are using and your individual details.
Setup
Create a directory for your analysis and save in it a script called set_dirs.sh which creates a variable called DATA which points to your dataset. The file should look something as follows:
# Run this file using "source set_dirs.sh" to set the following variable
DATA=/nobackup/proj/my_account/data
Download the apptainer container file kogwas.sif and save it in your analysis directory or somewhere accessible.
Create an hpc directory to save scripts to run the data preparation and analysis.
mkdir hpc
See Initial data preparation on how your data should be formatted.
Running scripts
Running a pipeline script that is stored inside the container is not too much different from before. Instead of running scripts like this:
../new_knockoffgwas_pipeline/run_lmm.sh 1 22 $DATA/mydata ...
they should be run like this:
apptainer exec --bind $DATA:/data kogwas.sif run_lmm.sh 1 22 /data/mydata ...
The $DATA variable should point to the location of your data on the host system. This path must be bind-mounted to the \data directory inside the container, since the scripts running in the container expect to find the data there.
The sub-sections below will take you through each step of using the pipeline with the container. As before, all scripts will need to be modified for your details and the requirements of the HPC machine that you are using.
Data preprocessing
Firstly, some map files need to be created with script hpc/pre_create_map_files.sh:
#!/bin/bash
#SBATCH --partition=default_free
#SBATCH --account=my_account
#SBATCH --cpus-per-task=1
#SBATCH --mem=20GB
#SBATCH --output=slurm_pre_create_map_files.out
# Load modules
module load apptainer
#Set dirs
source set_dirs.sh
date
echo "Running on $HOSTNAME PBC pre-analysis creating map files using container"
apptainer exec --bind $DATA:/data kogwas.sif run_pre_create_map_files.sh 1 22 /data/Nicola results /data/genetic_maps/genetic_map_GRCh37
date
Run it with:
sbatch hpc/pre_create_map_files.sh
Note that just the apptainer module needs to be loaded instead of various packages which are now handled by the container.
Next the data needs to be phased and the IBD segments calculated with script hpc/pre.sh:
#!/bin/bash
#SBATCH --partition=default_free
#SBATCH --account=my_account
#SBATCH --cpus-per-task=1
#SBATCH --mem=20GB
#SBATCH --array=1-22 # Tasks to run, corresponds to chromosome number
#SBATCH --output=slurm_pre_%a.out
# Load modules
module load apptainer
#Set dirs
source set_dirs.sh
date
echo "Running on $HOSTNAME PBC pre-analysis data preparing using container"
# Phase chromosome data
apptainer exec --bind $DATA:/data kogwas.sif run_pre_phasing.sh $SLURM_ARRAY_TASK_ID /data/mydata pbc results
# It may be necessary to change the segment length and window size until suitable IBD data is returned
apptainer exec --bind $DATA:/data kogwas.sif run_pre_ibd.sh $SLURM_ARRAY_TASK_ID /data/mydata pbc results 25 3
date
Run it with:
sbatch hpc/pre.sh
There may be some problems setting the segment length and window size, see Running data preprocessing for more details.
Analysis
See Step 2. Performing KnockOffGWAS for more details on running the KnockOffGWAS analysis. To do so using the container create a script hpc/analysis.sh something similar to the following:
#!/bin/bash
#SBATCH --partition=default_free
#SBATCH --account=my_account
#SBATCH --cpus-per-task=1
#SBATCH --mem=100GB
#SBATCH --array=1-22 # Run tasks for given chromosomes
#SBATCH --output=slurm_anal_%a.out
# Load modules
module load apptainer
# Set dirs
source ./set_dirs.sh
date
echo "Running on $HOSTNAME PBC analysis using container"
apptainer exec --bind $DATA:/data kogwas.sif run_knockof_gwas.sh $SLURM_ARRAY_TASK_ID /data/mydata pbc 0.1 results
date
and run it with:
sbatch hpc/analysis.sh
BOLT-LMM Analysis
See Step 3. BOLT-LMM Analysis for further details. Create an HPC script to do the preprocessing in the hpc directory called cont_bolt_lmm.sh which should look something like the following:
#!/bin/bash
#SBATCH --partition=default_free
#SBATCH --account=my_account
#SBATCH --cpus-per-task=1
#SBATCH --mem=10GB
#SBATCH --output=slurm_bolt_lmm.out
# Load modules
module load apptainer
# Set dirs
source ./set_dirs.sh
date
echo "Running on $HOSTNAME PBC BOLT-LMM analysis using container"
apptainer exec --bind $DATA:/data kogwas.sif run_lmm.sh 1 22 /data/mydata pbc lmm_results /data/tables/LDSCORE.1000G_EUR.GRCh38.tab.gz /data/tables/genetic_map_hg19_withX.txt.gz
date
As before, this will need to be updated for the requirements of the HPC machine that you are using.
Run the analysis as an array job on the HPC with the following command:
sbatch hpc/bolt_lmm.sh
or whatever is appropriate for the HPC machine you are using.
Results for the analysis will be stored in the results/lmm with a statistics file for chromosomes 1-22 named stats_chr1_chr22_lmm.txt along with two clumping files clump_chr1_chr22_lmm_regions.txt and clump_chr1_chr22_lmm_clumped.tab.
Run the analysis on the HPC with the following command:
sbatch hpc/bolt_lmm.sh
or whatever is appropriate for the HPC machine you are using.
Removing temporary files
A number of files are produced during the data preprocessing and during analysis calculations. These files can be removed by running a script, which also restores the .fam file which needed to have its format altered. If you remove the temporary files and wish to rerun the analysis for any reason you will need to rerun the data preprocessing scripts also.
The script, run_remove_temporary_files.sh, can be ran from the container as follows:
source set_dirs.sh
apptainer exec --bind $DATA:/data kogwas.sif run_remove_temporary_files.sh /data/mydata
Where the parameter is the path and filename of the data used in the analysis.
Visualisation
The visualisation of the results should still be done without the container using the shiny R app, app.R, which can be found here.
See Step 4. Visualisation for details.