Alphafold

Introduction

Alphafold is a protein structure prediction tool developed by DeepMind (Google). It uses a novel machine learning approach to predict 3D protein structures from primary sequences alone. The source code is available on Github. It has been deployed in all RCAC clusters, supporting both CPU and GPU.

It also relies on a huge database. The full database (~2.2TB) has been downloaded and setup for users.

Protein struction prediction by alphafold is performed in the following steps:

  • Search the amino acid sequence in uniref90 database by jackhmmer (using CPU)

  • Search the amino acid sequence in mgnify database by jackhmmer (using CPU)

  • Search the amino acid sequence in pdb70 database (for monomers) or pdb_seqres database (for multimers) by hhsearch (using CPU)

  • Search the amino acid sequence in bfd database and uniclust30 (updated to uniref30 since v2.3.0) database by hhblits (using CPU)

  • Search structure templates in pdb_mmcif database (using CPU)

  • Search the amino acid sequence in uniprot database (for multimers) by jackhmmer (using CPU)

  • Predict 3D structure by machine learning (using CPU or GPU)

  • Structure optimisation with OpenMM (using CPU or GPU)

For more information, please check:
Home page: https://github.com/deepmind/alphafold

Versions

  • 2.3.0

  • 2.3.1

Commands

  • run_alphafold.sh

Usage

The usage of Alphafold on our cluster is very straightford, users can create a flagfile containing the database path information:

run_alphafold.sh --flagfile=full_db.ff --fasta_paths=XX --output_dir=XX ...

Users can check its detaied user guide in its Github.

full_db_20230311.ff (for alphafold v3)

Example contents of full_db_20231031.ff for monomer:

--db_preset=full_dbs
--bfd_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt
--data_dir=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/
--uniref90_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/uniref90/uniref90.fasta
--mgnify_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/mgnify/mgy_clusters_2022_05.fa
--uniref30_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/uniref30/UniRef30_2021_03
--pdb70_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/pdb70/pdb70
--template_mmcif_dir=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/pdb_mmcif/mmcif_files
--obsolete_pdbs_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/pdb_mmcif/obsolete.dat
--hhblits_binary_path=/usr/bin/hhblits
--hhsearch_binary_path=/usr/bin/hhsearch
--jackhmmer_binary_path=/usr/bin/jackhmmer
--kalign_binary_path=/usr/bin/kalign

Example contents of full_db_20231031.ff for multimer:

--db_preset=full_dbs
--bfd_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt
--data_dir=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/
--uniref90_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/uniref90/uniref90.fasta
--mgnify_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/mgnify/mgy_clusters_2022_05.fa
--uniref30_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/uniref30/UniRef30_2021_03
--pdb_seqres_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/pdb_seqres/pdb_seqres.txt
--uniprot_database_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/uniprot/uniprot.fasta
--template_mmcif_dir=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/pdb_mmcif/mmcif_files
--obsolete_pdbs_path=/cluster/tufts/biocontainers/datasets/alphafold/db_20231031/pdb_mmcif/obsolete.dat
--hhblits_binary_path=/usr/bin/hhblits
--hhsearch_binary_path=/usr/bin/hhsearch
--jackhmmer_binary_path=/usr/bin/jackhmmer
--kalign_binary_path=/usr/bin/kalign

Example job using CPU

Warning

Using #!/bin/sh -l as shebang in the slurm job script will cause the failure of some biocontainer modules. Please use #!/bin/bash instead.

Note

Notice that since version 2.2.0, the parameter --use_gpu_relax=False is required.

To run alphafold using CPU:

#!/bin/bash
#SBATCH -p PartitionName  # batch or your group's own partition
#SBATCH -t 24:00:00
#SBATCH -N 1
#SBATCH -n 1
#SBATCH -c 10
#SBATCH --mem=64G
#SBATCH --job-name=alphafold
#SBATCH --mail-type=FAIL,BEGIN,END
#SBATCH --error=%x-%J-%u.err
#SBATCH --output=%x-%J-%u.out

module purge
module load alphafold/2.3.1

run_alphafold.sh --flagfile=full_db_20231031.ff  \
    --fasta_paths=sample.fasta --max_template_date=2022-02-01 \
    --output_dir=af2_full_out --model_preset=monomer \
    --use_gpu_relax=False

Example job using GPU

Warning

Using #!/bin/sh -l as shebang in the slurm job script will cause the failure of some biocontainer modules. Please use #!/bin/bash instead.

To run alphafold using GPU:

#!/bin/bash
#SBATCH -p PartitionName  # gpu or preempt
#SBATCH -t 24:00:00
#SBATCH -N 1
#SBATCH -n 1
#SBATCH -c 10
#SBATCH --mem=64G
#SBATCH --gres=gpu:1
#SBATCH --job-name=alphafold
#SBATCH --mail-type=FAIL,BEGIN,END
#SBATCH --error=%x-%J-%u.err
#SBATCH --output=%x-%J-%u.out

module purge
module load alphafold/2.3.1

run_alphafold.sh --flagfile=full_db_20231031.ff  \
    --fasta_paths=sample.fasta --max_template_date=2022-02-01 \
    --output_dir=af2_full_out --model_preset=monomer \
    --use_gpu_relax=True