AlphaFold

This package provides an implementation of the inference pipeline of AlphaFold v2. For simplicity, we refer to this model as AlphaFold throughout the rest of this document.

We also provide:

1. An implementation of AlphaFold-Multimer. This represents a work in progress and AlphaFold-Multimer isn't expected to be as stable as our monomer AlphaFold system. Read the guide for how to upgrade and update code.
2. The technical note containing the models and inference procedure for an updated AlphaFold v2.3.0.
3. A CASP15 baseline set of predictions along with documentation of any manual interventions performed.

Any publication that discloses findings arising from using this source code or the model parameters should cite the AlphaFold paper and, if applicable, the AlphaFold-Multimer paper.

Please also refer to the Supplementary Information for a detailed description of the method.

You can use a slightly simplified version of AlphaFold with this Colab notebook or community-supported versions (see below).

If you have any questions, please contact the AlphaFold team at alphafold@deepmind.com.

Installation and running your first prediction

You will need a machine running Linux, AlphaFold does not support other operating systems. Full installation requires up to 3 TB of disk space to keep genetic databases (SSD storage is recommended) and a modern NVIDIA GPU (GPUs with more memory can predict larger protein structures).

Please follow these steps:

1. Install Docker.

   • Install NVIDIA Container Toolkit for GPU support.
   • Setup running Docker as a non-root user.

2. Clone this repository and cd into it.

   bash git clone https://github.com/deepmind/alphafold.git cd ./alphafold

3. Download genetic databases and model parameters:

   • Install aria2c. On most Linux distributions it is available via the package manager as the aria2 package (on Debian-based distributions this can be installed by running sudo apt install aria2).

   • Please use the script scripts/download_all_data.sh to download and set up full databases. This may take substantial time (download size is 556 GB), so we recommend running this script in the background:

   bash scripts/download_all_data.sh <DOWNLOAD_DIR> > download.log 2> download_all.log &

   • Note: The download directory <DOWNLOAD_DIR> should not be a subdirectory in the AlphaFold repository directory. If it is, the Docker build will be slow as the large databases will be copied into the docker build context.

   • It is possible to run AlphaFold with reduced databases; please refer to the complete documentation.

4. Check that AlphaFold will be able to use a GPU by running:

   bash docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi

   The output of this command should show a list of your GPUs. If it doesn't, check if you followed all steps correctly when setting up the NVIDIA Container Toolkit or take a look at the following NVIDIA Docker issue.

   If you wish to run AlphaFold using Singularity (a common containerization platform on HPC systems) we recommend using some of the third party Singularity setups as linked in https://github.com/deepmind/alphafold/issues/10 or https://github.com/deepmind/alphafold/issues/24.

5. Build the Docker image:

   bash docker build -f docker/Dockerfile -t alphafold .

   If you encounter the following error:

   W: GPG error: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC E: The repository 'https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64 InRelease' is not signed.

   use the workaround described in https://github.com/deepmind/alphafold/issues/463#issuecomment-1124881779.

6. Install the run_docker.py dependencies. Note: You may optionally wish to create a Python Virtual Environment to prevent conflicts with your system's Python environment.

   bash pip3 install -r docker/requirements.txt

7. Make sure that the output directory exists (the default is /tmp/alphafold) and that you have sufficient permissions to write into it.

8. Run run_docker.py pointing to a FASTA file containing the protein sequence(s) for which you wish to predict the structure (--fasta_paths parameter). AlphaFold will search for the available templates before the date specified by the --max_template_date parameter; this could be used to avoid certain templates during modeling. --data_dir is the directory with downloaded genetic databases and --output_dir is the absolute path to the output directory.

   bash python3 docker/run_docker.py \ --fasta_paths=your_protein.fasta \ --max_template_date=2022-01-01 \ --data_dir=$DOWNLOAD_DIR \ --output_dir=/home/user/absolute_path_to_the_output_dir

9. Once the run is over, the output directory shall contain predicted structures of the target protein. Please check the documentation below for additional options and troubleshooting tips.

Genetic databases

This step requires aria2c to be installed on your machine.

AlphaFold needs multiple genetic (sequence) databases to run:

• BFD,
• MGnify,
• PDB70,
• PDB (structures in the mmCIF format),
• PDB seqres – only for AlphaFold-Multimer,
• UniRef30 (FKA UniClust30),
• UniProt – only for AlphaFold-Multimer,
• UniRef90.

We provide a script scripts/download_all_data.sh that can be used to download and set up all of these databases:

• Recommended default:

  bash scripts/download_all_data.sh <DOWNLOAD_DIR>

  will download the full databases.

• With reduced_dbs parameter:

  bash scripts/download_all_data.sh <DOWNLOAD_DIR> reduced_dbs

  will download a reduced version of the databases to be used with the reduced_dbs database preset. This shall be used with the corresponding AlphaFold parameter --db_preset=reduced_dbs later during the AlphaFold run (please see AlphaFold parameters section).

:ledger: Note: The download directory <DOWNLOAD_DIR> should not be a subdirectory in the AlphaFold repository directory. If it is, the Docker build will be slow as the large databases will be copied during the image creation.

We don't provide exactly the database versions used in CASP14 – see the note on reproducibility. Some of the databases are mirrored for speed, see mirrored databases.

:ledger: Note: The total download size for the full databases is around 556 GB and the total size when unzipped is 2.62 TB. Please make sure you have a large enough hard drive space, bandwidth and time to download. We recommend using an SSD for better genetic search performance.

:ledger: Note: If the download directory and datasets don't have full read and write permissions, it can cause errors with the MSA tools, with opaque (external) error messages. Please ensure the required permissions are applied, e.g. with the sudo chmod 755 --recursive "$DOWNLOAD_DIR" command.

The download_all_data.sh script will also download the model parameter files. Once the script has finished, you should have the following directory structure:

$DOWNLOAD_DIR/                             # Total: ~ 2.62 TB (download: 556 GB)
    bfd/                                   # ~ 1.8 TB (download: 271.6 GB)
        # 6 files.
    mgnify/                                # ~ 120 GB (download: 67 GB)
        mgy_clusters_2022_05.fa
    params/                                # ~ 5.3 GB (download: 5.3 GB)
        # 5 CASP14 models,
        # 5 pTM models,
        # 5 AlphaFold-Multimer models,
        # LICENSE,
        # = 16 files.
    pdb70/                                 # ~ 56 GB (download: 19.5 GB)
        # 9 files.
    pdb_mmcif/                             # ~ 238 GB (download: 43 GB)
        mmcif_files/
            # About 199,000 .cif files.
        obsolete.dat
    pdb_seqres/                            # ~ 0.2 GB (download: 0.2 GB)
        pdb_seqres.txt
    small_bfd/                             # ~ 17 GB (download: 9.6 GB)
        bfd-first_non_consensus_sequences.fasta
    uniref30/                              # ~ 206 GB (download: 52.5 GB)
        # 7 files.
    uniprot/                               # ~ 105 GB (download: 53 GB)
        uniprot.fasta
    uniref90/                              # ~ 67 GB (download: 34 GB)
        uniref90.fasta

bfd/ is only downloaded if you download the full databases, and small_bfd/ is only downloaded if you download the reduced databases.

Model parameters

While the AlphaFold code is licensed under the Apache 2.0 License, the AlphaFold parameters and CASP15 prediction data are made available under the terms of the CC BY 4.0 license. Please see the Disclaimer below for more detail.

The AlphaFold parameters are available from https://storage.googleapis.com/alphafold/alphafold_params_2022-12-06.tar, and are downloaded as part of the scripts/download_all_data.sh script. This script will download parameters for:

• 5 models which were used during CASP14, and were extensively validated for structure prediction quality (see Jumper et al. 2021, Suppl. Methods 1.12 for details).
• 5 pTM models, which were fine-tuned to produce pTM (predicted TM-score) and (PAE) predicted aligned error values alongside their structure predictions (see Jumper et al. 2021, Suppl. Methods 1.9.7 for details).
• 5 AlphaFold-Multimer models that produce pTM and PAE values alongside their structure predictions.

Updating existing installation

If you have a previous version you can either reinstall fully from scratch (remove everything and run the setup from scratch) or you can do an incremental update that will be significantly faster but will require a bit more work. Make sure you follow these steps in the exact order they are listed below:

1. Update the code.
   • Go to the directory with the cloned AlphaFold repository and run git fetch origin main to get all code updates.
2. Update the UniProt, UniRef, MGnify and PDB seqres databases.
   • Remove <DOWNLOAD_DIR>/uniprot.
   • Run scripts/download_uniprot.sh <DOWNLOAD_DIR>.
   • Remove <DOWNLOAD_DIR>/uniclust30.
   • Run scripts/download_uniref30.sh <DOWNLOAD_DIR>.
   • Remove <DOWNLOAD_DIR>/uniref90.
   • Run scripts/download_uniref90.sh <DOWNLOAD_DIR>.
   • Remove <DOWNLOAD_DIR>/mgnify.
   • Run scripts/download_mgnify.sh <DOWNLOAD_DIR>.
   • Remove <DOWNLOAD_DIR>/pdb_mmcif. It is needed to have PDB SeqRes and PDB from exactly the same date. Failure to do this step will result in potential errors when searching for templates when running AlphaFold-Multimer.
   • Run scripts/download_pdb_mmcif.sh <DOWNLOAD_DIR>.
   • Run scripts/download_pdb_seqres.sh <DOWNLOAD_DIR>.
3. Update the model parameters.
   • Remove the old model parameters in <DOWNLOAD_DIR>/params.
   • Download new model parameters using scripts/download_alphafold_params.sh <DOWNLOAD_DIR>.
4. Follow Running AlphaFold.

Using deprecated model weights

To use the deprecated v2.2.0 AlphaFold-Multimer model weights:

1. Change SOURCE_URL in scripts/download_alphafold_params.sh to https://storage.googleapis.com/alphafold/alphafold_params_2022-03-02.tar, and download the old parameters.
2. Change the _v3 to _v2 in the multimer MODEL_PRESETS in config.py.

To use the deprecated v2.1.0 AlphaFold-Multimer model weights:

1. Change SOURCE_URL in scripts/download_alphafold_params.sh to https://storage.googleapis.com/alphafold/alphafold_params_2022-01-19.tar, and download the old parameters.
2. Remove the _v3 in the multimer MODEL_PRESETS in config.py.

Running AlphaFold

The simplest way to run AlphaFold is using the provided Docker script. This was tested on Google Cloud with a machine using the nvidia-gpu-cloud-image with 12 vCPUs, 85 GB of RAM, a 100 GB boot disk, the databases on an additional 3 TB disk, and an A100 GPU. For your first run, please follow the instructions from [Installation and running your first prediction](#