AlphaFold 3 workflow

A Snakemake workflow for high-throughput AlphaFold 3 structure predictions

Switch to the 'parallel' branch of this repository if you plan to execute this workflow in an HPC environment that does not support 'consumable resources'.

🚀 What’s new?

📖 Better documentation to make setup & usage smoother
🔄 Support for different running modes, including:
    🧲 Pulldown
    💊 Virtual drug screening
    🔬 All-vs-all pairwise interactions
🛠️ Future plans: Adding stoichiometry screening support

TO DO

• Add steps for downstream analyses such as relaxation, assembly, binding site prediction, scoring etc.

Steps to setup & execute

1. Build the Singularity container

Install singularity. See here or here for instructions. Run the following command to build the Singularity container that supports parallel inference runs:

singularity build alphafold3_parallel.sif docker://ntnn19/alphafold3:latest_parallel_a100_40gb

Or

singularity build alphafold3_parallel.sif docker://ntnn19/alphafold3:latest_parallel_a100_80gb

Notes

Make sure to download the required AlphaFold3 databases and weights before proceeding.

2. Clone This repository

Clone this repository.

git clone https://github.com/ntnn19/AlphaFold3_workflow.git

Go to the repository location

cd AlphaFold3_workflow

An example JSON and CSV files are available in the example directory: example/example.json example/all_vs_all.csv example/pulldown.csv example/virtual_drug_screen.csv

3. Create & activate the Snakemake environment

Install mamba or micromamba if not already installed.

Then, set up and activate the environment using the following commands:

mamba env create -p $(pwd)/venv -f environment.yml
mamba activate $(pwd)/venv

For Maxwell users

module load maxwell mamba
. mamba-init
mamba env create -p $(pwd)/venv -f environment.yml
mamba activate $(pwd)/venv

Or if using micromamba

micromamba env create -p $(pwd)/venv -f environment.yml
eval "$(micromamba shell hook --shell=bash)"
micromamba activate $(pwd)/venv

4. Configure the workflow

Open config/config.yaml with your favorite text editor. Edit the values to your needs.

Mandatory workflow flags:

This workflow adapts the input preparation logic from AlphaFold3-GUI.

• output_dir: <path_to_your_output_directory> # Stores the outputs of this workflow
• af3_flags: # configures AlphaFold 3
  • af3_container: <path_to_your_alphafold3_container>
• input_csv: <path_to_your_csv_table>
• tmp_dir: <path_to_your_tmp_dir>

For running the default workflow, the user must provide a csv table such as the following:

default: example/default.csv

job_name	type	id	sequence
TestJob1	protein	A	MVLSPADKTNVKAAW
TestJob1	protein	B	MVLSPADKTNVKAAW
TestJob2	protein	C	MVLSPADKTNVKAAW

Explanation:

For explanation and full list of optional columns, see AlphaFold3-GUI api tutorial

Optional workflow flags:

The workflow supports running AlphaFold 3 in different modes: all-vs-all, pulldown, virtual-drug-screen, or stoichio-screen (TBD)

To run the workflow in a specific mode the user must provide the flag 'mode' in the config/config.yaml file.

For example:

input_csv: example/virtual_drug_screen_df.csv
output_dir: output
tmp_dir: tmp
mode: virtual-drug-screen
# n_splits: 4  # Optional, for running using the 'parallel' branch of this repo. To maximize resources utilization, the value of this flag should correspond to min(number_of_predictions, number_of_multi-GPU_nodes). 
af3_flags:
  --af3_container: alphafold3_parallel.sif

To run the workflow using a custom msa the user must provide the flag 'msa_option' in the config/config.yaml file and set it to 'custom', i.e. msa_option: custom. Setting 'msa_option' to 'custom' would require adding more columns to the input csv tables described below. For more information about this option, see here.

Examples for supported input_csv files for each mode:

all-vs-all: example/all_vs_all.csv

id	type	sequence
p1	rna	AUGGCA
p2	protein	MKPSFDR
p3	protein	MVLSPADKTNVKAAW

Explanation:

• id: A unique identifier for each entry.
• type: The biological macromolecule type (protein, dna, or rna).
• sequence: The nucleotide or amino acid sequence.

virtual drug screen: example/virtual_drug_screen.csv

id	type	sequence	drug_or_target	target_id	drug_id
p2	protein	MASEQASDTTVCIK	target	t1
l1	ligand	CC(=O)Oc1ccccc1C(=O)O	drug		d1
l2	ligand	CN1C=NC2=C1C(=O)N(C(=O)N2C)	drug		d1
p1	protein	MHIKPEERF	target	t1
p3	protein	ANHIREQDS	target	t2
l3	ligand	CN1C=NC2	drug		d2

Explanation:

• id: A unique identifier for each entry.
• type: The type of compound (protein, ligand, etc.).
• sequence: The amino acid sequence for proteins or the chemical structure for ligands.
• drug_or_target: Indicates whether the compound is a "drug" or "target".
• Optional columns:
  • target_id: The identifier for the target.
  • drug_id: The identifier for the drug.

The optional columns can be used to screen single drugs against multimeric targets, multiple drugs against monomeric targets, or multiple drugs against multimeric targets.

pulldown: example/pulldown.csv

id	type	sequence	bait_or_target	target_id	bait_id
p2	protein	MASEQASDTTVCIK	target	t2
b1	protein	MASEQASDTTVCIK	bait		b1
b2	protein	MASEQASDTTVCIK	bait		b2
p1	protein	MHIKPEERF	target	t1
p3	protein	ANHIREQDS	target	t2
b3	protein	ANHIREQDS	bait		b1

Explanation:

• id: A unique identifier for each entry.
• type: The biological macromolecule type (protein, dna, or rna).
• sequence: The nucleotide or amino acid sequence.
• bait_or_target: Indicates whether the protein is a "bait" or "target" in the experiment.
• Optional columns:
  • target_id: The identifier for the target.
  • bait_id: The identifier for the bait.

The optional columns can be used to pulldown multimeric targets with monomeric baits, multimeric targets with multimeric baits, or monomeric targets with multimeric baits.

Optional AlphaFold3 flags:

Include the optional flags within the scope of the af3_flags.

For flags explanations, run the following command:

singularity run <path_to_your_alphafold3_singularity_container> \
python /app/alphafold/run_alphafold.py --help

The optional flags are:

--buckets

--conformer_max_iterations

--flash_attention_implementation

--gpu_device

--hmmalign_binary_path

--hmmbuild_binary_path

--hmmsearch_binary_path

--jackhmmer_binary_path

--jackhmmer_n_cpu

--jax_compilation_cache_dir

--max_template_date

--mgnify_database_path

--nhmmer_binary_path

--nhmmer_n_cpu

--ntrna_database_path

--num_diffusion_samples

--num_recycles

--num_seeds

--pdb_database_path

--rfam_database_path

--rna_central_database_path

--save_embeddings

--seqres_database_path

--small_bfd_database_path

--uniprot_cluster_annot_database_path

--uniref90_database_path

Example for a config/config.yaml file with optional AlphaFold 3 flags:

input_csv: example/virtual_drug_screen_df.csv
output_dir: output
tmp_dir: tmp
mode: virtual-drug-screen 
af3_flags:
  --af3_container: alphafold3_parallel.sif
  --num_diffusion_samples: 5

5. Configure the profile (Optional)

For running this workflow on HPC using slurm, you can modify profile/config.yaml to make it compatible with your HPC setting. More details can be found here.

6. Run the workflow

Information on snakemake flags can be found here

Dry run

python workflow/scripts/prepare_workflow.py config/config.yaml
snakemake -s workflow/Snakefile \
-j unlimited -c all \
-p -k -w 30 --rerun-triggers mtime -n --configfile config/config.yaml

Local run

python workflow/scripts/prepare_workflow.py config/config.yaml
snakemake -s workflow/Snakefile \
--use-singularity --singularity-args  \
'--nv -B <your_alphafold3_weights_dir>:/root/models -B <your_output_dir>:/root/af_output -B <your_alphafold3_databases_dir>:/root/public_databases -B <your_alphafold3_tmp_dir>:/tmp --env XLA_CLIENT_MEM_FRACTION=3.2' \
-j unlimited -c all \
-p -k -w 30 --rerun-triggers mtime --configfile config/config.yaml

slurm run

python workflow/scripts/prepare_workflow.py config/config.yaml
snakemake -s workflow/Snakefile \
--use-singularity --singularity-args  \
'--nv -B <your_alphafold3_weights_dir>:/root/models -B <your_output_dir>:/root/af_output -B <your_alphafold3_databases_dir>:/root/public_databases -B <your_alphafold3_tmp_dir>:/tmp --env XLA_CLIENT_MEM_FRACTION=3.2' \
-j unlimited -c all \
-p -k -w 30 --rerun-triggers mtime --workflow-profile profile --configfile config/config.yaml

If you find this useful, please consider giving it a star! ⭐