#!/bin/bash

#SBATCH -J <name of your job>
#SBATCH -o <desired name for log file>.%j.log
#SBATCH -A <account>
#SBATCH -p <m2_gpu|deeplearning>
#SBATCH --gres=gpu:1 # NOTE: AlphaFold is multi gpu capable, but
                     #       apparently not stable.
#SBATCH -c 8         # NOTE: Non-GPU components of AlphaFold are
                     #       hardly able to use more than 8 CPUs.
#SBATCH --mem=20G    # NOTE: For really large protein complexes more
                     #       memory might be needed.
#SBATCH -t 300       # NOTE: This is plenty of time for small and medium
                     #       sized problems. Increase the time value in case
                     #       of bigger simulations.


################################################################################
# load environment

module purge
module load bio/AlphaFold

################################################################################
# variables

INFILE=<path to input FASTA file>
# NOTE: AlphaFold per default creates an output file using the input file name.
#       In order to avoid overriding old runs, you can indicate an own, holding
#       the unique jobid.
OUTDIR=$PWD/alphafold_test_$SLURM_JOB_ID
mkdir OUTDIR

# NOTE: As the environment variable $ALPHA_FOLD_DATA is set by the module,
#       no further data indicators are required upon starting the programm.
srun alphafold \
    --output_dir=$OUTDIR \
    --fasta_paths=$INFILE \
    --max_template_date=<max_template_date, e.g. '2020-05-14'> \
    --db_preset=<full_dbs|reduced_dbs> \
    --model_preset=<monomer|multimer>

#!/bin/bash

#SBATCH -J <name of your job>
#SBATCH -o <desired name for log file>.%j.log
#SBATCH -A <account>
#SBATCH -p <m2_gpu|deeplearning>
#SBATCH --gres=gpu:1 # NOTE: AlphaFold is multi gpu capable, but
                     #       apparently not stable.
#SBATCH -c 8         # NOTE: Non-GPU components of AlphaFold are
                     #       hardly able to use more than 8 CPUs.
#SBATCH --mem=20G    # NOTE: For really large protein complexes more
                     #       memory might be needed.
#SBATCH -t 300       # NOTE: This is plenty of time for small and medium
                     #       sized problems. Increase the time value in case
                     #       of bigger simulations.


################################################################################

module purge
module load tools/AppTainer # NOTE: The AppTainer module provides support for
                            #       a Singularity container.

################################################################################

INFILE=<path to input FASTA file>
# NOTE: AlphaFold per default creates an output file using the input file name.
#       In order to avoid overriding old runs, you can indicate an own, holding
#       the unique jobid.
OUTDIR=$PWD/alphafold_test_$SLURM_JOB_ID
mkdir -p $OUTDIR

ALPHAFOLD_DATA_DIR=/lustre/project/alphafold_users
CONTAINERPATH=${ALPHAFOLD_DATA_DIR}/container

# NOTE: The $ALPHAFOLD_DATA_DIR is not accepted by the current container version, therefore
#       the individual flags need to be defined.

srun singularity run --env TF_FORCE_UNIFIED_MEMORY=1,XLA_PYTHON_CLIENT_MEM_FRACTION4.0,OPENMM_CPU_THREADS=8
  -B .:/etc --nv ${CONTAINERPATH}/<alphafold_container_version>.sif
  --fasta_paths=$INFILE
  --output_dir=$OUTDIR
  --max_template_date=<max_template_date, e.g. '2020-05-14'>
  --data_dir=$ALPHAFOLD_DATA_DIR
  --uniref90_database_path=$ALPHAFOLD_DATA_DIR/uniref90/uniref90.fasta
  --mgnify_database_path=$ALPHAFOLD_DATA_DIR/mgnify/mgy_clusters_2018_12.fa
  --small_bfd_database_path=$ALPHAFOLD_DATA_DIR/small_bfd/bfd-first_non_consensus_sequences.fasta
  --pdb70_database_path=$ALPHAFOLD_DATA_DIR/pdb70/pdb70
  --template_mmcif_dir=$ALPHAFOLD_DATA_DIR/pdb_mmcif/mmcif_files
  --obsolete_pdbs_path=$ALPHAFOLD_DATA_DIR/pdb_mmcif/obsolete.dat
  --use_gpu_relax
  --db_preset=<full_dbs|reduced_dbs>
  --model_preset=<monomer|multimer>