AlphaFold-3

AlphaFold 3

How to use AlphaFold3 with the new chorus GPU nodes.

Reference data is available here: /shared/biodata/alphafold3

Model Parameters

You will need to download your own personal copy of the parameter data. The request form has been sent to Hutch Legal Console for approval. I will download and share when the paper work is done. To request access to the AlphaFold 3 model parameters, please complete this form.

Sbatch Example

Example sbatch script to run AlphaFold3. Save a copy of this and modify it to suit your data.

• --cpus==8 is the current max we are working on method to request more. If you ask for more your job will stay in pending state.
• --gpus=2 is a serving suggestion. Request 2 or 4 to run AlphaFold3.
• Please update the --job-name for your work.
• This script works from the Rhino’s or maestro
• You need to define OUTPUT_DIR, JSON_PATHS, and MODEL_PATH

#!/bin/sh
#SBATCH --job-name="af3"
#SBATCH --partition=chorus
#SBATCH --nodes=1
#SBATCH --cpus-per-task=8
#SBATCH --mem=1024GB
#SBATCH --gpus=2
 
# User defined locations
export OUTPUT_DIR=$SCRATCH/alphafoldtest/output
export JSON_PATHS=alphafold_input.json
 
# Obtaining Model Parameters
# Contact AlphaFold to get access to model parameters
#
export MODEL_PATH=/Path/To/Models
 
export DOWNLOAD_DIR=/shared/biodata/alphafold3
SIF=/app/software/AlphaFold/containers/alpafold3.sif
 
apptainer exec \
  --nv \
  --bind $DOWNLOAD_DIR:/root/public_databases \
  $SIF \
  python /app/alphafold/run_alphafold.py \
  --json_path=$JSON_PATHS \
  --model_dir=$MODEL_PATH \
  --output_dir=$OUTPUT_DIR