AlphaFold 3

AlphaFold is an AI system developed by Google DeepMind that predicts a protein's 3D structure from its amino acid sequence. It regularly achieves accuracy competitive with experiment. - https://alphafold.ebi.ac.uk/

AlphaFold 3 is provided on M3 via the alphafold/3.0.0 module.

We provide:

• All of the reference datasets for AlphaFold 3,
• The AlphaFold 3 inference pipeline.

We do not provide the model weights, so you must obtain these for yourself. This is done to ensure you understand the legal limitations of using this model, specifically the Terms Of Use. If your project has, or may have, commercial interests we recommend you seek legal advice.

Read our Access instructions for details on accessing AlphaFold 3 on M3.

Access instructions

1. Request access to the alphafold3 software in Karaage at https://hpc.erc.monash.edu.au/karaage/software/66/. See Modules - Licensed software if you are unsure about this.
2. Request your own AlphaFold 3 model weights according to https://github.com/google-deepmind/alphafold3?tab=readme-ov-file#obtaining-model-parameters.
3. Once you can access the model weights, copy them over to M3.

Usage

Make sure you have followed the Access instructions before trying to use AlphaFold 3 on M3.

Before you load the module, you must set ALPHAFOLD3_MODEL_WEIGHTS to the path of your own AlphaFold 3 model weights.

export ALPHAFOLD3_MODEL_WEIGHTS=/path/to/your/alphafold3_weights
module load alphafold/3.0.0

tip

You can add the line:

export ALPHAFOLD3_MODEL_WEIGHTS=/path/to/your/alphafold3_weights

to your ~/.bashrc so the environment variable is automatically set every time you login to M3.

You should now have access to run_alphafold.py. An example command is:

run_alphafold.py \
    --json_path=/path/to/fold_input.json \
    --output_dir=/path/to/af_output

Here are some M3-specific notes:

• In our M3 alphafold/3.0.0 module, run_alphafold.py calls apptainer under the hood.
• We automatically call --model-dir=$ALPHAFOLD3_MODEL_WEIGHTS for you, so you don't need to specify that.
• We automatically set --nv for you so you get GPU support.
• We automatically expose key M3 filepaths into the underlying container, meaning you should be able to use normal M3 filepaths (not weird filepaths like /root/af_input/ as you might see in the AlphaFold 3 docs).