Set of scripts to run STRUCTURE in parallel on computing grids like MetaCentrum. Scripts are designed for grids and clusters using PBS Pro, but can be easily adopted for another queue system.
Version: 1.3
Vojtěch Zeisek, https://trapa.cz/.
See https://github.com/V-Z/structure-multi-pbspro, ask about usage or so at https://github.com/V-Z/structure-multi-pbspro/discussions and report any issues or wishes using https://github.com/V-Z/structure-multi-pbspro/issues.
GNU General Public License 3.0, see LICENSE.md and https://www.gnu.org/licenses/gpl-3.0.html.
This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
STRUCTURE itself process single file in time. It has simple Java GUI available to create batch task and run on desktop, or also possibly on MetaCentrum. Other option is ParallelStructure R package (see my older example and slides --- chapter Structure from slide 204), but it has problems with some input file formats. It runs on single computer, using multiple cores.
Provided scripts distribute individual runs of STRUCTURE among multiple nodes (computers, servers) in computing cluster/grid, which speeds up everything a lot. STRUCTURE must be rune repeatedly for different K, and also for each K repeatedly. Here, structure_multi_1_submitter.sh will submit each single run as individual computing job. As soon as structure_multi_1_submitter.sh is done, the computing jobs are submitted to the queue and user can monitor their progress by commands like qstat (depending on scheduling system of the computing grid/cluster), and see if something already appeared in the output directory.
The scripts are written for Linux servers. They might be running on another UNIX systems. Apart of BASH, the only requirement is STRUCTURE. It is already installed on MetaCentrum (module structure/2.3.4), so that user can simply load the module. If using own installation of STRUCTURE, either comment out or update respective line in script structure_multi_2_qsub.sh. If you are unsure how to work in Linux command line on computing cluster, consult e.g. my slides or MetaCentrum documentation.
Either download and decompress latest release or clone the Git repository:
git clone https://github.com/V-Z/structure-multi-pbspro.git
cd structure-multi-pbspro/
./structure_multi_1_submitter.sh -h
./structure_multi_1_submitter.sh -vConsider copying of both scripts structure_multi_1_submitter.sh and structure_multi_2_qsub_run.sh into some folder dedicated to store scripts and software like ~/bin/ to have them available in PATH.
Prepare input file and MAINPARAMS and EXTRAPARAMS files according to STRUCTURE manual. The scripts need them as input. They then overwrite the K stated in MAINPARAMS to get the range of Ks, and name outputs according to K and repetition.
Script structure_multi_1_submitter.sh will use qsub to submit multiple jobs to calculate individual STRUCTURE runs. E.g. for K ranging from 1 to 10 and with 10 repetitions it will submit 100 jobs, which can be by cluster/grid computed in parallel (queueing system will decide according to cluster load).
-h--- Print help and exit.-v--- Print script version, author and license and exit.-s--- Path to STRUCTURE binary. If not provided, it must be available inPATHvariable.-m--- Path to STRUCTURE MAINPARAMS file. Consult STRUCTURE manual (required).-e--- Path to STRUCTURE EXTRAPARAMS file. Consult STRUCTURE manual (required).-i--- Input data file. Consult STRUCTURE manual (required).-n--- Output files base name. If not provided, default isres. It can contain only Latin characters, numbers, dots, underscores or dashes. The name for each output file will be asres.k.X.rep.Y.out, whereXis actual K andYis repetition.-o--- Output directory. Should be empty. If provided directory does not exist, it will be created (required).-f--- Minimal K. Default is 1.-k--- Maximal K. Default is 10.-r--- How many times run for each K. Default is 10.-w--- Walltime (maximal running time) in hours for individual job to finish. Default is 24. See documentation of your cluster/grid scheduling system (e.g. MetaCentrum).
If -s is not specified (and structure_multi_1_submitter.sh doesn't find it in PATH), then structure_multi_2_qsub_run.sh will load module structure/2.3.4 and use it.
Script structure_multi_1_submitter.sh will pass needed variables --- i.e. input files, output name and directory, path to STRUCTURE binary (if needed) and particular K and repetition --- to structure_multi_2_qsub_run.sh which will do the calculation. The latter script uses variables passed via qsub from script structure_multi_1_submitter.sh and calculates single run of STRUCTURE. As all the jobs are submitted in single step, the cluster queueing system can highly parallelize all calculations (if the cluster has enough performance, all jobs can dun in parallel).
If there is no need to edit the scripts (see following chapter), typical usage is like this:
# Submit the job
./structure_multi_1_submitter.sh -m mainparams.txt -e extraparams.txt -i input.str -n myanalysis -o str_outdir -f 1 -k 10 -r 10
# Monitor if jobs are running
# Different command might be needed on computing grids/clusters using different scheduling system
qstat -w -n -1 -u "${USER}" -xIf jobs are correctly submitted, but there are no *_f output files in the output directory (str_outdir in the above example), check output logs as there is probably something wrong with your input data. If so, consult STRUCTURE manual. If input files were prepared on Windows, ensure they have correct EOL, e.g. using dos2unix.
Normally, there is no need to manually run structure_multi_2_qsub_run.sh --- it is used by structure_multi_1_submitter.sh. It is possible to use structure_multi_2_qsub_run.sh for single run (see comments in it what to edit in such case).
Reduced microsatellite (SSRs) dataset of Nuphar lutea from Czech rivers (65 individuals, 5 loci, 13 populations) from 3 disconnected river basins (Elbe, Morava, Oder). Input file nuphar_in.str, STRUCTURE MAINPARAMS file nuphar_mainparams.txt and STRUCTURE EXTRAPARAMS file nuphar_extraparams.txt.
# Clone the git repository
git clone https://github.com/V-Z/structure-multi-pbspro.git
# Go to the directory
cd structure-multi-pbspro/
# See content
ls
# See help
./structure_multi_1_submitter.sh -h
# Submit the jobs
./structure_multi_1_submitter.sh -m nuphar_mainparams.txt -e nuphar_extraparams.txt -i nuphar_in.str -n Nuphar -o nuphar_str_out -f 1 -k 15 -r 10 -w 1
# Monitor how the jobs are running...
qstat -w -n -1 -u "${USER}" -x
# From time to time see results in the output directory...
ls -lh nuphar_out/Edits might be required on clusters/grids using different scheduling system than PBS Pro. Of course, improvements are welcomed, but edit the code only if you know what you are doing. ;-)
If your cluster/grid is using different scheduling system than PBS on MetaCentrum, edit in last section of structure_multi_1_submitter.sh the qsub line. Also, if you need to submit the job to particular queue, change time to run, needed memory or so (e.g. for larger data), edit required resources on that qsub line.
If your cluster/grid is using different method to cleanup of temporal (scratch) directories than MetaCentrum, edit or remove trap commands in structure_multi_2_qsub_run.sh. If your cluster/grid is using different method to manage application modules than MetaCentrum (there structure/2.3.4), edit or remove the block with module add command in structure_multi_2_qsub_run.sh. If your cluster/grid is using different name of variable pointing to temporal working directory than SCRATCH on MetaCentrum, replace all occurrences of SCRATCH by the correct variable name in structure_multi_2_qsub_run.sh.
Of course, improvements, generalizations for easier work on another clusters/grids are welcomed. :-)
For next step collect all res.k.X.rep.Y.out_f files in the output directory. Select the best K using e.g. Structure_sum R script (see my older example and slides --- chapter Structure from slide 204) or Structure Harvester. Structure_sum was originally written by Dorothée Ehrich and updated for modern R by me --- see https://github.com/V-Z/structure-sum --- and Marek Šlenker. Align and reorder the results with CLUMPP and draw final plots by e.g. distruct. See also my older complete example.