Skip to content

AlphaFold 3

We provide AlphaFold v3.0.3 on the HPC for use by approved users.

Note

AlphaFold 3 requires an NVIDIA GPU with a Compute Capability 8.0 or greater, which includes the Ampere, Ada Lovelace, Hopper, and Blackwell architectures.

Approval process

We restrict access to the AlphaFold 3 model parameters to users who have accepted their terms of service. To become an approved user, follow these steps:

  1. Complete this Google form.
  2. Within 3 business days, you should receive an email from alphafold@google.com that starts with "I'm writing to confirm that you have been granted access to the AlphaFold 3 model parameters subject to the AlphaFold 3 Model Parameters Terms of Use."
  3. Forward this email to divilovk@oregonstate.edu
  4. CQLS staff will manually add you to a user group with access to the model parameters and will respond to your email stating that you have been approved.
  5. Once approved, you can immediately run the command below to run AlphaFold 3. We provide an HPC-wide model parameters file so you don't need to use your personal file.

Using AlphaFold 3 on the HPC

We provide AlphaFold v3.0.3 on the HPC as singularity containers located in /local/cqls/singularity/images/. Both x86 and ARM containers are provided.

The command to run the x86 container is:

singularity run --nv \
  --bind /local/cqls/db \
  /local/cqls/singularity/images/alphafold_v3.0.3_x86.sif \
  --model_dir=/local/cqls/db/alphafold3_params \
  --db_dir=/local/cqls/db/alphafold3 \
  --json_path=input/fold_input.json \
  --output_dir=output

This command assumes that in your working directory you have an input file fold_input.json that is in the input folder and the output will be sent to the output folder.

Adjust these variables to your preference. Change alphafold_v3.0.3_x86.sif to alphafold_v3.0.3_arm.sif to run the ARM container.

To see the full or short help menus, run singularity run --nv /local/cqls/singularity/images/alphafold_v3.0.3_x86.sif with the --helpfull or --helpshort flags.