alphafold

Module Name

alphafold

Last Updated On

08/29/2023

Support Level

Secondary Support

Software Access Level

Open Access

Home Page

https://github.com/deepmind/alphafold

Software Description

Alphafold is an AI-based protein structure inference pipeline developed by DeepMind, which can predict a protein\'s 3D structure from its amino acid sequence.

Slurm Example

Example scripts for submission to k40 and v100 partitions are provided below. Note that you will need to customize these scripts with your email (replacing sample_email@umn.edu), specific project directory (replacing project_dir), and input/output file locations (replacing ./output and ./input/my_fasta.fasta).

k40:

#!/bin/bash -l        
#SBATCH --time=24:00:00
#SBATCH --ntasks=24
#SBATCH --mem=20g
#SBATCH --tmp=100g
#SBATCH -p k40
#SBATCH --gres=gpu:k40:2
#SBATCH --mail-type=ALL  
#SBATCH --mail-user=sample_email@umn.edu

module load alphafold/2.0

cd project_dir

ln -s $alphafold_path alphafold  2>/dev/null || true

run_alphafold.sh -o ./output -m model_1 -f ./input/my_fasta.fasta -t 2020-05-14

v100:

#!/bin/bash -l        
#SBATCH --time=24:00:00
#SBATCH --ntasks=24
#SBATCH --mem=20g
#SBATCH --tmp=100g
#SBATCH -p v100
#SBATCH --gres=gpu:v100:2
#SBATCH --mail-type=ALL  
#SBATCH --mail-user=sample_email@umn.edu
module load alphafold/2.0

cd project_dir

ln -s $alphafold_path alphafold  2>/dev/null || true

run_alphafold.sh -o ./output -m model_1 -f ./input/my_fasta.fasta -t 2020-05-14

v100 multimer workflow:

#!/bin/bash -l        
#SBATCH --time=24:00:00
#SBATCH --ntasks=24
#SBATCH --mem=20g
#SBATCH --tmp=100g
#SBATCH -p v100
#SBATCH --gres=gpu:v100:2
#SBATCH --mail-type=ALL  
#SBATCH --mail-user=sample_email@umn.edu
module load alphafold/2.1.1-multimer

cd project_dir

ln -s $alphafold_path alphafold  2>/dev/null || true

run_alphafold.sh -o ./output -m multimer -f ./input/my_multimer_fasta.fasta -t 2021-11-01

General Linux

You can load alphafold for use in a Linux environment via:

module load alphafold

Running alphafold at MSI is a little different than what is indicated in DeepMind\'s original documentation. Since we are using a non-docker implementation based on https://github.com/kalininalab/alphafold_non_docker , we need to adapt our workflow accordingly. A workflow might look like:

module load alphafold

# We need a symlink to this directory wherever we are running alphafold, since several reference files are 
# supposed to be located in ./alphafold 
cd project_dir 
ln -s $alphafold_path alphafold 2>/dev/null || true

# Use a provided wrapper script to invoke the pipeline 
run_alphafold.sh -o ./output_directory -m model_1 -f ./input/my_fasta.fasta -t 2020-05-14

Default

2.3.1-multimer

Other Modules

2.0, 2.1.1-multimer, 2.3.1-multimer

Default

2.3.1-multimer

Other Modules

2.0, 2.1.1-multimer, 2.3.1-multimer

Default

2.3.1-multimer

Other Modules

2.0, 2.1.1-multimer, 2.3.1-multimer