AlphaFold

AlphaFold predicts the three-dimensional structure of a protein from its amino acid sequence, work that otherwise depends on slow, resource-intensive experimental methods such as X-ray crystallography or cryo-EM. Version 3 goes beyond single chains: it also predicts how proteins associate with DNA, RNA, small-molecule ligands, and ions (Abramson et al., Nature 2024), which makes it useful for studying molecular complexes and binding interactions.

The current module on the cluster is alphafold/3.0.3. Access is provided through Apptainer containers built for GPU computation.

Prerequisites

Obtain model parameters from Google:

https://forms.gle/svvpY4u2jsHEwWYS6

Set the environment variable to your model directory:

export MODEL_DIR=/home/$USER/alphafold3_models/

Setup

Create Working Directories

mkdir -p ~/alphafold
mkdir -p /resnick/scratch/$USER/alphafold/out

Load Module

module load alphafold/3.0.3

Copy Example Files

cp /resnick/software9/external/alphafold/examples/test.json ~/alphafold/
cp /resnick/software9/external/alphafold/examples/alphafold3.sub ~/

Job Submission

Example SLURM Script

#!/bin/bash
#SBATCH --job-name=alphafold_run
#SBATCH --time=1-00:00:00
#SBATCH --partition=gpu
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --gres=gpu:h100:1
#SBATCH --cpus-per-task=16
#SBATCH --mem=32G
#SBATCH [email protected]
#SBATCH --mail-type=ALL

export MODEL_DIR=/home/$USER/alphafold3_models/
module load alphafold/3.0.3

time alphafold \
    --output_dir=/resnick/scratch/$USER/alphafold/out \
    --json_path=/home/$USER/alphafold/test.json

Submit

sbatch alphafold3.sub

Output

  • Results appear in the specified --output_dir

  • Job logs: slurm-[jobid].out

Additional Options

View all available flags:

alphafold --helpshort

Tips

  • Use H100 GPUs for best performance

  • Allocate sufficient memory (32 GB minimum recommended)

  • Output to scratch storage for large predictions

  • Monitor jobs with squeue -u $USER