Boltz 2.2.1

Webpage

https://github.com/jwohlwend/boltz

Version

2.2.1

Installation Environment

• Python 3.12.12 (pyenv 2.6.17)

Brief Installation Procedure

$ module load pyenv/2.6.17
$ export PYENV_ROOT=/apl/pyenv/repos/2.6.17
$ pyenv install 3.12.12
$ pyenv virtualenv 3.12.12 3.12.12-boltz-2.2.1
$ pyenv activate 3.12.12-boltz-2.2.1
(3.12.12-boltz-2.2.1) $ eval "$(pyenv init -)"
(3.12.12-boltz-2.2.1) $ pip install boltz[cuda]==2.2.1 -U

Job Sample

The workflow involves first running on the login node (ccfep), accessing the Colabfold server to perform MSA and data processing, then executing the prediction on the GPU node with the generated job script.

#!/bin/sh

INPUT=prot.yaml
JOBFILE=sample-prot-jsub.sh
MSALOG=msa-prot.logfile

NCPUS=16
NTHREADS=8
NGPUS=1
WALLTIME="16:00:00"

BOLTZ_VER="2.2.1"

# ------------------------------------------------

module -s purge
module -s load boltz/${BOLTZ_VER}

# generate MSA on ccfep
boltz predict ${INPUT} --use_msa_server >& ${MSALOG}

# prediction
cat -<< EOF > ${JOBFILE}
#!/bin/sh
#PBS -l select=1:ncpus=${NCPUS}:mpiprocs=1:ompthreads=${NTHREADS}:ngpus=${NGPUS}
#PBS -l walltime=${WALLTIME}
cd \$PBS_O_WORKDIR
module -s purge
module -s load boltz/${BOLTZ_VER}
boltz predict ${INPUT}
EOF
jsub ${JOBFILE}

Notes

• This is installed in the pyenv environment prepared on the shared area.
• Prediction failed to run if GPU is not available. (First "boltz predict")
• If processed data is already available, it will be used to make predictions. (Second "boltz predict")