Alphafold3

AlphaFold3 Module Guide

Overview

This guide provides instructions on using the AlphaFold3 module in an HPC environment with Slurm and Apptainer (Singularity).

Loading the Module

To use AlphaFold3, first load the module:

module load alphafold3

This will automatically load the required Apptainer module and set up necessary environment variables.

Environment Variables

After loading the module, the following environment variables are available:

• $AF3_CONTAINER - Path to the AlphaFold3 Singularity container
• $AF3_MODELS - Path to AlphaFold3 model parameters
• $AF3_DB - Path to the AlphaFold3 database
• $AF3_SRC - Path to AlphaFold3 source directory

Understanding Bind Mounts in Singularity

Singularity requires **bind mounts** to provide external directories inside the container. This ensures that the container can access necessary files without modifying its internal structure.

A bind mount follows this format:

--bind /path/on/host:/path/inside/container

- `/path/on/host` → The folder on your actual system (outside the container) - `/path/inside/container` → Where this folder will be accessible **inside** the container

For example:

--bind /home/user/data:/root/input_data

This allows the **`/home/user/data`** directory on the host system to appear inside the container at **`/root/input_data`**.

Binding a Folder for Your Input Data

The default AlphaFold3 example from Google DeepMind uses `/root/af_input`, but this is **not a required folder inside the container**. You can bind your input folder to **any directory** inside the container.

For example, if your input JSON is located at `/home/user/alphafold_inputs/fold_input.json`, you can bind it inside the container at **any location**:

--bind /home/user/alphafold_inputs:/root/custom_folder

Then, the JSON file should be referenced as:

--json_path=/root/custom_folder/fold_input.json

Refer to the official AlphaFold3 documentation for details on how to structure the input file: [AlphaFold3 Input File Guide](https://github.com/google-deepmind/alphafold3/blob/main/docs/input.md)

Listing All Available Command Flags

To see a full list of available command-line options for AlphaFold3, run the following inside the container:

singularity exec --nv \
    --bind $AF3_SRC:/root/custom_folder \
    --bind /tmp:/root/af_output \
    --bind $AF3_MODELS:/root/models \
    --bind $AF3_DB:/root/public_databases \
    --bind /home/user/alphafold_inputs:/root/custom_folder \
    $AF3_CONTAINER \
    python /root/custom_folder/run_alphafold.py --helpfull

This will provide detailed documentation on all possible flags and parameters that can be used when running AlphaFold3.

Running AlphaFold3

To run AlphaFold3 inside the Singularity container, use:

singularity exec --nv \
    --bind $AF3_SRC:/root/custom_folder \
    --bind /tmp:/root/af_output \
    --bind $AF3_MODELS:/root/models \
    --bind $AF3_DB:/root/public_databases \
    --bind /home/user/alphafold_inputs:/root/custom_folder \
    $AF3_CONTAINER \
    python /root/custom_folder/run_alphafold.py \
        --json_path=/root/custom_folder/fold_input.json \
        --model_dir=/root/models \
        --db_dir=/root/public_databases \
        --output_dir=/root/af_output

Replace `/home/user/alphafold_inputs` with the actual path to your input folder.

Binding a Folder That is Not Related to Built-In Environment Variables

If you need to bind a folder that is **not already covered by** `$AF3_SRC`, `$AF3_DB`, or `$AF3_MODELS`, you can specify it manually using the `--bind` option. Example:

--bind /external/data:/root/custom_data

This makes `/external/data` accessible inside the container at `/root/custom_data`.

Then, reference files from inside the container:

--json_path=/root/custom_data/my_input.json

Submitting an AlphaFold3 Job to Slurm

To submit an AlphaFold3 job to Slurm, create a script (e.g., `alphafold3_job.sh`):

#!/bin/bash
#SBATCH --job-name=alphafold3
#SBATCH --partition=gpu-general
#SBATCH --output=alphafold3.out
#SBATCH --error=alphafold3.err
#SBATCH --gres=gpu:1,af3
#SBATCH --cpus-per-task=8
#SBATCH --mem=64G
#SBATCH --time=1-00:00:00

module load alphafold3

singularity exec --nv \
    --bind $AF3_SRC:/root/custom_folder \
    --bind /tmp:/root/af_output \
    --bind $AF3_MODELS:/root/models \
    --bind $AF3_DB:/root/public_databases \
    --bind /home/user/alphafold_inputs:/root/custom_folder \
    $AF3_CONTAINER \
    python /root/custom_folder/run_alphafold.py \
        --json_path=/root/custom_folder/fold_input.json \
        --model_dir=/root/models \
        --db_dir=/root/public_databases \
        --output_dir=/root/af_output

Submit the job using:

sbatch alphafold3_job.sh

Unloading the Module

To unload the module when done:

module unload alphafold3

This will also unload the Apptainer module.

Troubleshooting

• If the module does not load, ensure the correct name is used: `module avail alphafold3`
• If Slurm fails to allocate a node, check available resources: `sinfo -o "%N %G"`
• If the container fails to run, verify that the paths in `$AF3_CONTAINER`, `$AF3_MODELS`, and `$AF3_DB` are correct.
• If input files are not found, ensure the correct directory is bound and referenced properly inside the container.