AlphaFold
This is not a complete article: This is a draft, a work in progress that is intended to be published into an article, which may or may not be ready for inclusion in the main wiki. It should not necessarily be considered factual or authoritative.
This package provides an implementation of the inference pipeline of AlphaFold v2.0. This is a completely new model that was entered in CASP14 and published in Nature. For simplicity, we refer to this model as AlphaFold throughout the rest of this document.
Any publication that discloses findings arising from using this source code or the model parameters should cite the AlphaFold paper.
The source code of this package can be found in their GitHub page along with some documentation.
Usage in Compute Canada systems
As you might have seen from their documentation, they explain the usage via Docker. In Compute Canada we do not provide Docker as container, but singularity (see our documentation at https://docs.computecanada.ca/wiki/Singularity). However, we have created a wheel to use AlphaFold in a python environment.
AlphaFold in Python environment
1. AlphaFold has a number of non-python dependencies that need to be loaded ahead of time. For example, cuda, kalign, hmmer, and openmm. Luckily all these dependencies are available though our stack:
[name@cluster ~]$ module load gcc/9 openmpi cuda/11.1 cudnn/8.2.0 kalign/2.03 hmmer/3.2.1 openmm-alphafold/7.5.1 hh-suite/3.3.0 python/3.8
2. Clone the AlphaFold repository:
[name@cluster ~]$ git clone https://github.com/deepmind/alphafold.git
3. Then you can proceed to create the python virtual environment and activate it by:
[name@cluster ~]$ virtualenv --no-download ~/my_env && source ~/my_env/bin/activate
3. Now you can install AlphaFold and its dependencies by:
(my_env)[name@cluster ~]$ pip install --no-index six==1.15 numpy==1.19.2 scipy==1.4.1 pdbfixer alphafold
Now AlphaFold is ready to be used. Note that to use AlphaFold ouside a container, you need to use the run_alphafold.py script that is provided in the repository.
Creating the virtual environment in the jobscript
As you probably have read in Creating_and_using_a_virtual_environment, you can also take advantage of the local installs on compute nodes:
#!/bin/bash
#SBATCH --job-name=alphafold_run
#SBATCH --account=def-someprof # adjust this to match the accounting group you are using to submit jobs
#SBATCH --time=0-03:00 # adjust this to match the walltime of your job
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --gres=gpu:1 # You need to request one GPU to be able to run AlphaFold properly
#SBATCH --cpus-per-task=1 # adjust this if you are using parallel commands
#SBATCH --mem=4000 # adjust this according to the memory requirement per node you need
#SBATCH --mail-user=you@youruniversity.ca # adjust this to match your email address
#SBATCH --mail-type=ALL
# Load your modules as before
module load gcc/9 openmpi cuda/11.1 cudnn/8.2.0 kalign/2.03 hmmer/3.2.1 openmm-alphafold/7.5.1 hh-suite/3.3.0 python/3.8
# Generate your virtual environment in $SLURM_TMPDIR
virtualenv --no-download ${SLURM_TMPDIR}/my_env && source ${SLURM_TMPDIR}/my_env/bin/activate
# Install alphafold and dependencies
pip install --no-index six==1.15 numpy==1.19.2 scipy==1.4.1 pdbfixer
# Run your commands
python /path/to/repo/run_alphafold.py --help
Remember to provide the /path/to/repo/ when using run_alphafold.py
Databases
Note that AlphaFold requires a set of datasets/databases that need to be downloaded. Also notice that we prefer you avoid using `aria2c`. For your convenience we have included a modified version of the scripts within the AlphaFold wheel, so within the virtual environment you can do:
(my_env)[name@cluster ~]$ download_all_data.sh path/to/download
Note that this might take a while and SHOULD NOT BE DONE IN THE COMPUTE NODES. Instead, you should use the data transfer nodes or the login nodes. Since the download might take a while we recommend you do this in a screen or Tmux sessions. If your path/to/download is stored in `$DOWNLOAD_DIR`, then the structure of your data should be:
$DOWNLOAD_DIR/ # Total: ~ 2.2 TB (download: 428 GB)
bfd/ # ~ 1.8 TB (download: 271.6 GB)
# 6 files.
mgnify/ # ~ 64 GB (download: 32.9 GB)
mgy_clusters.fa
params/ # ~ 3.5 GB (download: 3.5 GB)
# 5 CASP14 models,
# 5 pTM models,
# LICENSE,
# = 11 files.
pdb70/ # ~ 56 GB (download: 19.5 GB)
# 9 files.
pdb_mmcif/ # ~ 206 GB (download: 46 GB)
mmcif_files/
# About 180,000 .cif files.
obsolete.dat
uniclust30/ # ~ 87 GB (download: 24.9 GB)
uniclust30_2018_08/
# 13 files.
uniref90/ # ~ 59 GB (download: 29.7 GB)
uniref90.fasta
This is important when passing the commands to AlphaFold.
Running AlphaFold
Once you have everything setup, you can run a production run of AlphaFold by:
#!/bin/bash
#SBATCH --job-name=alphafold_run
#SBATCH --account=def-someprof # adjust this to match the accounting group you are using to submit jobs
#SBATCH --time=0-03:00 # adjust this to match the walltime of your job
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --gres=gpu:1 # You need to request one GPU to be able to run AlphaFold properly
#SBATCH --cpus-per-task=8 # adjust this if you are using parallel commands
#SBATCH --mem=32G # adjust this according to the memory requirement per node you need
#SBATCH --mail-user=you@youruniversity.ca # adjust this to match your email address
#SBATCH --mail-type=ALL
# Set the path to download dir
DOWNLOAD_DIR=/projects/def-someprof/${USER}/data # Set the appropriate path to your downloaded data
DATA_DIR=/projects/def-someprof/${USER}/input # Set the appropriate path to your supporting data
# Load your modules as before
module load gcc openmpi cuda/11.1 cudacore/.11.1.1 cudnn/8.2.0 kalign hmmer hh-suite openmm python/3.7
# Generate your virtual environment in $SLURM_TMPDIR
virtualenv --no-download ${SLURM_TMPDIR}/my_env && source ${SLURM_TMPDIR}/my_env/bin/activate
# Install alphafold and dependencies
pip install --no-index six==1.15 numpy==1.19.2 scipy==1.4.1 pdbfixer alphafold
# Run your commands
python run_alphafold.py --bfd_database_path ${DOWNLOAD_DIR}/bfd --data_dir ${DATA_DIR} \
--fasta_paths ${DATA_DIR}/fasta1.fasta,${DATA_DIR}/fasta2.fasta,${DATA_DIR}/fasta3.fasta \
--hhblits_binary_path ${EBROOTHHMINSUITE}/bin/hhblits \
--hhsearch_binary_path ${EBROOTHHMINSUITE}/bin//hhsearch')
--jackhmmer_binary_path ${EBROOTHMMER}/bin/jackhmmer \
--kalign_binary_path ${EBROOTKALIGN}/bin/kalign
--mgnify_database_path ${DOWNLOAD_DIR}/mgnify --model_names model1,model2 \ # use the actual models you want
--output_dir ~/scratch/alphafold_output --pdb70_database_path ${DOWNLOAD_DIR}/pdb70 \
--template_mmcif_dir ${DATA_DIR}/Templates --uniclust30_database_path ${DOWNLOAD_DIR}/uniclust30
--uniref90_database_path ${DOWNLOAD_DIR}/uniref90
Using singularity
In case you want to try the conternarized version (NOT our preferred option), you can build a singularity container:
[name@cluster ~]$ module load singularity [name@cluster ~]$ singularity build alphafold.sif docker://uvarc/alphafold:2.0.0
Before trying to build it or run it, check our singularity documentation as there are particularities of each system that need to be taken into account.