High-Performance Computing at the NIH
GitHub YouTube @nih_hpc RSS Feed
Frealign

Frealign is a program for high-resolution refinement of 3D reconstructions from cryo-EM images of single particles.

References:

There are multiple versions of Frealign available. An easy way of selecting the version is to use modules. To see the modules available, type

module avail Frealign

To select a module, type

module load Frealign/[ver]

where [ver] is the version of choice.

Environment variables set:

The command frealign_v9_mp.exe is multithreaded. The number of threads is set by the $NCPUS environment variable. This variable is set dynamically by the Frealign module using the $SLURM_CPUS_ON_NODE variable set by the Slurm batch system. It can be overridden by setting $NCPUS prior to loading the Frealign module.
To see the value of $NCPUS prior to running, type 'module show Frealign'.

On Helix

To run Frealign, first create an mparameters file, edit it, then run (for example, frealign_run_refine):

module load Frealign
frealign_template
nano mparameters
frealign_run_refine
Batch job on Biowulf

Single Node Job

There are two ways of running a batch job on Biowulf. If your commands involve subtasks that run very quickly (for example reconstruction steps that last a few minutes each), it is much more efficient to run on a single node, using the local scratch disk. This is done by editing the mparameters file. In this case, because the path to the local scratch disk space is unknown prior to the job submission, the mparameters file contains a dummy tag "XXXXXXXX".

cluster_type         none
nprocessor_ref       16
nprocessor_rec       16
scratch_dir          /lscratch/XXXXXXXX

Create a batch input file (e.g. frealign.sh) which cds to the directory containing the mparameters file and substitutes the actual path to local disk into the dummy tag. For example:

#!/bin/bash
module load Frealign
cd /path/to/directory/where/mparameters/lives/
sed -i "s#/lscratch/XXXXXXXX#/lscratch/$SLURM_JOB_ID#" mparameters
frealign_run_refine

Submit this job using the Slurm sbatch command, allocating as many cpus as indicated in the nprocessor_ref and/or nprocessor_rec values, as well as local scratch space:

sbatch --cpus-per-task=16 --gres=lscratch:50 frealign.sh

Multiple Node Job

In cases where the subtasks of Frealign as expected to run for a significantly long time (more than 10 minutes), it may be better to distribute the subtasks across multiple nodes. Again, edit the mparameters file to indicate 'slurm' as the cluster type, and set nprocessor_{ref,rec} to no more than 50. In this case, you must use a shared scratch space, so either leave scratch_dir blank to indicate the current working directory, or set the value to a specific directory. The value for mp_cpus can be safely set to 2, because each sbatch job will allocate a minimum of 2 cpus.

cluster_type         slurm
nprocessor_ref       50
nprocessor_rec       50
mp_cpus              2

Create a batch script (for example frealign.sh):

#!/bin/bash
module load Frealign
cd /path/to/directory/where/mparameters/lives/
frealign_v9_mp.exe

Then submit to the cluster without any special allocations:

sbatch frealign.sh

Job Control via mparameters File

When the cluster_type is set to 'slurm', there are two values in mparameters which can give control the allocations for batch submission: qsub_string_ref and qsub_string_rec. These can be used to select specific partitions, allocations, etc. For example, to run subtasks on a different partition:

qsub_string_ref      "--partition=quick"
qsub_string_rec      "--partition=quick""

If the number of cpus used during reconstruction can be larger than 2, then both mp_cpus and qsub_string_rec need to complement each other. For example, to set then number of cpus to 8:

mp_cpus              8
qsub_string_rec      "--cpus-per-task=8"
Interactive job on Biowulf

Once an interactive node has been allocated,

sinteractive --cpus-per-task=32 --gres=lscratch:50

launch the job as you would on helix, making sure that the number of processors indicated in the mparameters file is equal to the number allocated, and that the cluster_type is set to 'none'.

Documentation