Run GAPSTOP™

Folder structure

To use the provided tm_param.star without any significant changes you should have the following folder structure:

tm_tutorial/
├── microtubule/
│   ├── microtubule.em
|   ├── mask_microtubule.em
├── angles_5_c13.txt
├── wedge_list.star
├── tm_param.star
└── 126_b4.mrc

The only parameter that has to be changed is the rootdir parameter that should contain a valid path to the tm_tutorial folder (including the tm_tutorial).

Note that the folder can contain more files so you do not need to delete anything from that folder.

It is not necessary to create the folder for the outputs - if it does not exist prior running the TM it will be create on the fly.

Run TM

Depending on your local installation you might need to activate the environment where the GAPSTOP™ is installed:

# assuming the environment is called gapstop
source /path/to/the/gapstop/bin/activate

Once you have all the inputs prepared and you set the correct paths you can run the TM directly:

gapstop run_tm -n 8 tm_param.star

The “n” parameter specifies the number of GPUs on which the TM should run.

Run TM on SLURM cluster

Here is an example of the bash script to run the GAPSTOP™ on a SLURM cluster - please note that some parameters such as “constraint” and “gres” are specific to your SLURM cluster setup and you should use/change them accordingly.

Example of tm_submit.sh script

#!/bin/bash -l
#SBATCH -o log_file
#SBATCH -e err_file
#SBATCH -D /path/to/working/directory/
#SBATCH -J tm_microtubule
#SBATCH --time=02:00:00
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=4
#SBATCH --cpus-per-task=18
#SBATCH --constraint="gpu"
#SBATCH --gres=gpu:a100:4
#SBATCH --gpu-bind=verbose,per_task:1

export OMP_NUM_THREADS=$SLURM_CPUS_PER_TASK
export OMP_PLACES=cores

source /path/to/the/gapstop/bin/activate
srun gapstop run_tm -n 8 tm_param.star

In this particular setup there is 4 GPUs per node and thus the total number of tiles is 8. This has to be adapted based on the SLURM specific setup (e.g in case there are only 2 GPUs per node the –ntasks-per-node will be 2 and the “n” should be either 4 or the number of nodes needs to be increased to 4 by setting –nodes=4).

To submit such script on SLURM cluster run:

chmod +x tm_submit.sh
sbatch tm_submit.sh

Results

The results will be stored in the tm_outputs folder: scores_0_126.em, angles_0_126.em and 0.log. See the section on results_eval <results_eval.html> to find out out how to proceed with their analysis.