Application Guide: AlphaFold 3¶

Note

This page provides details on AlphaFold 3. For information on previous versions of AlphaFold on BlueBEAR, please see:
https://bear-apps.bham.ac.uk/applications/AlphaFold/

Process Overview¶

Obtain the databases and models
Submit a job using the provided Apptainer image

The information on this page is based on the following AlphaFold 3 documentation:
https://github.com/google-deepmind/alphafold3/blob/main/README.md#installation-and-running-your-first-prediction

Obtaining the AlphaFold 3 Databases and Models¶

Model Parameters¶

Request access to the AlphaFold 3 model parameters, as per the instructions here:
https://github.com/google-deepmind/alphafold3/blob/main/docs/installation.md#obtaining-model-parameters
Info
The model parameters download size is approx. 970 MB.
Once your access has been granted, please download and store the data in an appropriate directory, taking care to ensure that the Model Parameters Terms of Use are not contravened.

Genetic Databases¶

Due to their size (approx. 630 GB), we have already downloaded the genetic databases and put them in a read-only location on the Research Data Store.
In order to run AlphaFold 3, these databases need to be stored in a writeable location so we suggest copying them to an appropriate RDS project directory (N.B. they won’t fit in a user home directory).

Please see the following batch script as a suggested method for doing this:

Example

#!/bin/bash

#SBATCH --ntasks 1
#SBATCH --time 5:0:0  # 5 hours

SOURCE_PATH="${BB_APPS_DATA}/AlphaFold/AlphaFold3/AlphaFold3_Genetic_Databases_20241126"

# Modify this path to the directory where you want the data to reside
DEST_PATH="/rds/projects/m/my_project/AF3_gen_DB"

rsync -rlptv --progress "${SOURCE_PATH}" "${DEST_PATH}"

N.B. if the copy doesn’t complete within the time limit, resubmit the script and it will resume from where it failed.

Submitting a job using the Apptainer image¶

AlphaFold 3 container image

We have built an Apptainer container image based on the Dockerfile that is referenced in the following AlphaFold 3 documentation section:
https://github.com/google-deepmind/alphafold3/blob/main/docs/installation.md#building-the-docker-container-that-will-run-alphafold-3

Once the AlphaFold 3 data is downloaded (see above), write your AlphaFold 3 input json file and then use the following batch script to submit it to BlueBEAR:

#!/bin/bash

#SBATCH --ntasks=1
# (6)!

############

MODEL_PRMS=/path/to/model/parameters # (1)!
GENETIC_DB=/rds/projects/m/my_project/AF3_gen_DB # (2)!

FOLD_JSON_INPUT=/path/to/json/input/file # (5)!
OUTPUT_DIR=/path/to/output

############

mkdir -p "${OUTPUT_DIR}"

CONTAINER_IMAGE=/rds/bear-apps/container-images/singularity/x86_64/a/AlphaFold3/alphafold3-3.0.0.sif # (3)!

apptainer exec \
     --nv \ # (4)!
     "${CONTAINER_IMAGE}" \
     python run_alphafold.py \
     --model_dir="${MODEL_PRMS}" \
     --db_dir="${GENETIC_DB}" \
     --json_path="${FOLD_JSON_INPUT}" \
     --output_dir="${OUTPUT_DIR}"

Path to the directory in which the model parameters are stored.
Path to the directory in which the genetic databases are stored. This should match the path in the $DEST_PATH variable in the above script.

To list the available versions of the container, please run:

ls -l /rds/bear-apps/container-images/singularity/x86_64/a/AlphaFold3

Only include --nv if running the job on a GPU node.
AlphaFold 3 input json file information.
Set the appropriate #SBATCH headers.