NBIS project id #6005
- 2022-01-21
Workflow will be restarted on Rackham using the config/myconfig_R-samples.yaml
configfile. However, the time limit should be updated on the pfam_scan rule as
it seems to be taking longer than expected for a few assemblies.
The config/cluster.yaml file should be updated to contain:
pfam_scan:
time: 14400The work directory should probably be unlocked as well. Then the workflow can be
restarted with (remember to remove the -n flag):
snakemake --profile slurm --use-conda --configfile config/myconfig_R-samples.yaml -j 100 -rpk -n annotate- November - December 2021
Project was set up on Rackham under /crex/proj/snic2020-6-126/nobackup/nbis-meta-2.3.2.
The nbis-meta workflow (v. 2.3.2) was downloaded from GitHub. The conda environment
was installed under /crex/proj/snic2020-6-126/nobackup/envs/nbis-meta using
command:
mkdir envs
conda env create -f environment.yml -p envs/nbis-metaConfig and sample files were set up as config/myconfig.yaml and sample_list_user-ids.tsv,
respectively.
Preprocessing
Preprocessing was run with cutadapt as:
snakemake --profile slurm --use-conda --configfile config/myconfig.yaml -j 100 -rpk -n qcAssembly
Assembly was run with megahit as:
snakemake --profile slurm --use-conda --configfile config/myconfig.yaml -j 100 -rpk -n assembleAnnotation
After this, an attempt at annotating the 'R' samples was made using a configfile
config/myconfig_R-samples.yaml:
snakemake --profile slurm --use-conda --configfile config/myconfig_R-samples.yaml -j 100 -rpk -n annotateHowever, the *.bam files generated by the workflow used up ~7TB of disk space
causing the project to hit the uquota limit around Dec 28, 2021. This also caused
the workflow to fail as files could no longer be moved from local node storage
to the project directory.