AlphaFold3

AlphaFold3 is protein structural prediction software from Google DeepMind and Isomorphic Labs. It is the successor to previous versions of Alphafold, which are available on HPCC.

Terms of Use and Model Parameters

While AlphaFold3 is distributed under a ShareAlike license, the model parameters are subject to terms of use agreement, particularly in regards to commercial applications and training other models.

To obtain a personal copy of the of AlphaFold3 model parameters, please request one from Google using the following form.

Alternatively, access to an institutional copy can be obtained by requesting access through the ICER contact form and acknowedgeling the terms of use there.

AlphaFold3 installed in HPCC

Available versions of AlphaFold3 can be found using the following command:

module avail AlphaFold3

After loading the module, you can run AlphaFold3 using the 'run_alphafold.py' wrapper script. For example:

module purge
module load AlphaFold3/3.0.1-foss-2023a-CUDA-12.4.0

run_alphafold.py \
     --json_path=protein_input.json \
     --model_dir=<path_to_model_parameters_file> \
     --db_dir=<path_to_alphafold3_database> \
     --output_dir=output \

In addition to the module, HPCC also provides a copy of the AlphaFold3 database at:

/mnt/research/common-data/alphafold/database_3/

However, users must supply their own model parameters file or gain access to the institutional copy.

Additional documentation on AlphaFold3 can be found at the AlphaFold3 GitHub.

Known Issues

AlphaFold3 requires a number of python libraries which currently cannot be built to operate on both avx512 and other node architectures at the same time.

ICER is aware of this issues and working on a solution.

Alternatives

While HPCC does not support Docker images, user can build or obtain their own Singualirty image which is compatible with HPCC.

ICER currently does not maintain its own Singularity image or AlphaFold3 nor makes guarantee of support for 3rd party images.