AlphaFold

AlphaFold

See FASRC Docs

Running AlphaFold Examples

We recommend running AlphaFold on GPU partitions because it runs much faster than solely using CPUs -- due to AlphaFold's GPU optimization. See slurm partitions for the specifics of each partition.

Below you will find a slurm script example run_alphafold.sh that uses the fasta file 5ZE6_1.fasta. This example assumes that run_alphafold.sh and my_fasta are located in the same directory. If they are located in different directories, you will have to edit my_fasta_path.

You will have to edit in the run_alphafold.sh script:

• SBATCH directives to suit your needs (e.g. time -t, number of cores -c, amount of memory --mem)
• my_fasta to suit to your own fasta file
• (optional) my_output_dir if you would like your output to go somewhere else
• (optional) my_fasta_path

Note: AlphaFold screen output goes to the stderr file (.err) rather than the stdout file (.out).

Monomer batch job

This example takes about 1 hour to run on Cannon in the gpu partition with 8 cores (-c 8).

Slurm script

User_Codes/Applications/AlphaFold/run_alphafold.sh

Lines 1 to 48 in d9c3f2f

	#!/bin/bash
	#SBATCH -J AF_monomer # Job name
	#SBATCH -p gpu # Partition(s) (separate with commas if using multiple)
	#SBATCH --gres=gpu:1 # number of GPUs
	#SBATCH -c 8 # Number of cores
	#SBATCH -t 03:00:00 # Time (D-HH:MM:SS)
	#SBATCH --mem=60G # Memory
	#SBATCH -o AF_mono_%j.out # Name of standard output file
	#SBATCH -e AF_mono_%j.err # Name of standard error file
	# set fasta file name
	# NOTE: assumes this is in the directory you are running this script in
	# note that you can run multiple proteins _sequentially_ (with the same model type)
	# the names need to be provided as "protein1.fasta,protein2.fasta"
	# if running multimer, provide one multifasta file
	# indicate oligomeric state by including extra copies of a sequence
	# they still require different _names_ though
	my_fasta=5ZE6_1.fasta
	# create and set path of output directory
	my_output_dir=output
	mkdir -p $my_output_dir
	# set model type (monomer, multimer, monomer_casp14, monomer_ptm)
	# see notes under fasta file if running multimer
	my_model_type=monomer
	# max pdb age
	# use if you want to avoid recent templates
	# format yyyy-mm-dd
	my_max_date="2100-01-01"
	# run AlphaFold monomer using Singularity
	singularity run --nv --env TF_FORCE_UNIFIED_MEMORY=1,XLA_PYTHON_CLIENT_MEM_FRACTION=4.0,OPENMM_CPU_THREADS=$SLURM_CPUS_PER_TASK,LD_LIBRARY_PATH=/usr/local/cuda-11.1/targets/x86_64-linux/lib/ --bind /n/holylfs04-ssd2/LABS/FAS/alphafold_database:/data /n/singularity_images/FAS/alphafold/alphafold_2.3.1.sif \
	--data_dir=/data/ \
	--bfd_database_path=/data/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt \
	--db_preset=full_dbs \
	--fasta_paths=$my_fasta \
	--max_template_date=$my_max_date \
	--mgnify_database_path=/data/mgnify/mgy_clusters_2022_05.fa \
	--model_preset=$my_model_type \
	--obsolete_pdbs_path=/data/pdb_mmcif/obsolete.dat \
	--output_dir=$my_output_dir \
	--pdb70_database_path=/data/pdb70/pdb70 \
	--template_mmcif_dir=/data/pdb_mmcif/mmcif_files \
	--uniref30_database_path=/data/uniref30/UniRef30_2021_03 \
	--uniref90_database_path=/data/uniref90/uniref90.fasta \
	--use_gpu_relax=True

Fasta file

User_Codes/Applications/AlphaFold/5ZE6_1.fasta

Lines 1 to 2 in f15ac0e

	>5ZE6_1
	MNLEKINELTAQDMAGVNAAILEQLNSDVQLINQLGYYIVSGGGKRIRPMIAVLAARAVGYEGNAHVTIAALIEFIHTATLLHDDVVDESDMRRGKATANAAFGNAASVLVGDFIYTRAFQMMTSLGSLKVLEVMSEAVNVIAEGEVLQLMNVNDPDITEENYMRVIYSKTARLFEAAAQCSGILAGCTPEEEKGLQDYGRYLGTAFQLIDDLLDYNADGEQLGKNVGDDLNEGKPTLPLLHAMHHGTPEQAQMIRTAIEQGNGRHLLEPVLEAMNACGSLEWTRQRAEEEADKAIAALQVLPDTPWREALIGLAHIAVQRDR

Multimer batch job

This example takes about 1-2 hours to run on Cannon in the gpu partition with 8 cores (-c 8).

Slurm script

User_Codes/Applications/AlphaFold/run_alphafold_multi.sh

Lines 1 to 60 in 1f69e32

	#!/bin/bash
	#SBATCH -J AF_multimer # Job name
	#SBATCH -p gpu # Partition(s) (separate with commas if using multiple)
	#SBATCH --gres=gpu:1 # number of GPUs
	#SBATCH -c 8 # Number of cores
	#SBATCH -t 03:00:00 # Time (D-HH:MM:SS)
	#SBATCH --mem=60G # Memory
	#SBATCH -o AF_multi_%j.out # Name of standard output file
	#SBATCH -e AF_multi_%j.err # Name of standard error file
	# set fasta file name
	# NOTE: assumes this is in the directory you are running this script in
	# note that you can run multiple proteins _sequentially_ (with the same model type)
	# the names need to be provided as "protein1.fasta,protein2.fasta"
	# if running multimer, provide one multifasta file
	# indicate oligomeric state by including extra copies of a sequence
	# they still require different _names_ though
	my_fasta=T1083_T1084.fasta
	# set fasta-specific subfolder and filepath
	# handling different possible .fasta suffixes
	fasta_name="${my_fasta//.fasta}"
	fasta_name="${fasta_name//.faa}"
	fasta_name="${fasta_name//.fa}"
	mkdir -p $fasta_name
	cp $my_fasta $PWD/$fasta_name
	my_fasta_path=$PWD/$fasta_name/$my_fasta
	# create and set path of output directory
	my_output_dir=af2_out
	mkdir -p $PWD/$fasta_name/$my_output_dir
	my_output_dir_path=$PWD/$fasta_name/$my_output_dir
	# set model type (monomer, multimer, monomer_casp14, monomer_ptm)
	# see notes under fasta file if running multimer
	my_model_type=multimer
	# max pdb age
	# use if you want to avoid recent templates
	# format yyyy-mm-dd
	my_max_date="2100-01-01"
	# run AlphaFold multimer using Singularity
	singularity run --nv --env TF_FORCE_UNIFIED_MEMORY=1,XLA_PYTHON_CLIENT_MEM_FRACTION=4.0,OPENMM_CPU_THREADS=$SLURM_CPUS_PER_TASK,LD_LIBRARY_PATH=/usr/local/cuda-11.1/targets/x86_64-linux/lib/ --bind /n/holylfs04-ssd2/LABS/FAS/alphafold_database:/data -B .:/etc --pwd /app/alphafold /n/singularity_images/FAS/alphafold/alphafold_2.3.1.sif \
	--data_dir=/data/ \
	--bfd_database_path=/data/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt \
	--db_preset=full_dbs \
	--fasta_paths=$my_fasta_path \
	--max_template_date=$my_max_date \
	--mgnify_database_path=/data/mgnify/mgy_clusters_2022_05.fa \
	--model_preset=$my_model_type \
	--obsolete_pdbs_path=/data/pdb_mmcif/obsolete.dat \
	--output_dir=$my_output_dir_path \
	--template_mmcif_dir=/data/pdb_mmcif/mmcif_files \
	--uniref30_database_path=/data/uniref30/UniRef30_2021_03 \
	--uniref90_database_path=/data/uniref90/uniref90.fasta \
	--pdb_seqres_database_path=/data/pdb_seqres/pdb_seqres.txt \
	--uniprot_database_path=/data/uniprot/uniprot.fasta \
	--use_gpu_relax=True

Fasta file

User_Codes/Applications/AlphaFold/T1083_T1084.fasta

Lines 1 to 4 in 1f69e32

	>T1083
	GAMGSEIEHIEEAIANAKTKADHERLVAHYEEEAKRLEKKSEEYQELAKVYKKITDVYPNIRSYMVLHYQNLTRRYKEAAEENRALAKLHHELAIVED
	>T1084
	MAAHKGAEHHHKAAEHHEQAAKHHHAAAEHHEKGEHEQAAHHADTAYAHHKHAEEHAAQAAKHDAEHHAPKPH

Submitting a slurm batch job that runs AlphaFold

Log in to Cannon (see login instructions). Go to the directory where run_alphafold.sh is located. Then submit a slurm batch job with the command:

# monomer job
sbatch run_alphafold.sh

# multimer job
sbatch run_alphafold_multi.sh