Openfold Non Docker Setup

OpenFold non-docker setup

Installation

First, you need to install anaconda (miniconda and mamba are fine as well, I'll stick to conda in the following). For a detailed instruction on how to install conda, please refer to their website. In the inference part we furthermore assume that anaconda3 is installed in the home directory.

The following installation steps are also covered in install.sh (fast way). For information on how to use this, please refer to it's help (./install.sh -h).

Storage size

Apart from the datasets that need to be downloaded, some memory is needed for the code and the weights. As of August 2022, all alphafold weights have a size of 11GB, all openfold weights have a size of 3.5GB. The code, including aws, needs 383MB.

TL;DR If you want to use

• openfold only, you need around 4GB of storage
• alphafold only, you need around 12GB of storage
• openfold and alphafold, you need around 15GB of storage

Download databases

To download all the databases needed for this is crucial for execution of openfold. There are two options to download the databases:

• Option 1: Use our download_db.sh script which uses wget, rsync, gunzip and tar instead of aria2c Our script maintains the AF2 download directory structure. To download the script, use wget:

  wget https://github.com/kalininalab/openfold_non_docker/blob/main/download_db.sh

  To get information on how to use it, call download_db.sh -h.

  To download all databases in full_dbs mode

  ./download_db.sh -d </home/johndoe/alphafold_data>

  To download the reduced version of the databases in reduced_dbs mode type

  ./download_db.sh -d </home/johndoe/alphafold_data> -m reduced_dbs

• Option 2: Follow https://github.com/deepmind/alphafold#genetic-databases

Installation - Fast way

Clone this repository and execute install.sh and run.sh to get the predictions.

git clone git@github.com:kalininalab/openfold_non_docker.git openfold
cd openfold
./install.sh -a -o -p <where/to/store/the/weights>

# run with alphafold weights
./run.sh -d <path/to/databases> -f <path/to/fastas> -c <path/to/conda> -s <path/to/save> -a <path/to/af_weights>

# run with openfold weights
./run.sh -d <path/to/databases> -f <path/to/fastas> -c <path/to/conda> -s <path/to/save> -o <model_name> <path/to/af_weights>

Installation - Slow way

We will install everything without sudo rights into a folder called openfold:

mkdir openfold
cd openfold

Setup Conda Environment

Create a conda environment and install some necessary packages

conda create --name openfold python=3.7
conda activate openfold
conda install -y -q -c conda-forge -c bioconda kalign2=2.04 hhsuite=3.3.0
pip install -q ml-collections==0.1.0 PyYAML==5.4.1 biopython==1.79

Access AWS

This is only necessary if you plan to use the retrained OpenFold parameters. This model also works with the original AlphaFold parameters, so this step is optional.

curl https://awscli.amazonaws.com/awscli-exe-linux-x86_64.zip -o awscliv2.zip
unzip -qq awscliv2.zip
./aws/install -i ./aws/ -b ./aws/
rm awscliv2.zip

The executable aws file is now located in ./v2/2.7.20/bin/.

Install OpenFold

git clone https://github.com/aqlaboratory/openfold
cd openfold
conda env create --file environment.yml
conda activate openfold_venv
pip install .
cd openfold/resources
wget https://git.scicore.unibas.ch/schwede/openstructure/-/raw/7102c63615b64735c4941278d92b554ec94415f8/modules/mol/alg/src/stereo_chemical_props.txt
cd ../../..

Download the weights

AlphaFold parameters

mkdir -p params/af_params
cd params/af_params
wget https://storage.googleapis.com/alphafold/alphafold_params_2022-01-19.tar
tar --extract --verbose --file alphafold_params_2022-01-19.tar --preserve-permissions
cd ../..

OpenFold Weights

To be able to execute the following lines of code, you need to follow the AWS steps above.

mkdir -p params/of_params
path/to/aws s3 cp --no-sign-request --region us-east-1 s3://openfold/openfold-params ./params/of_params/ --recursive

Inference

For inference with OpenFold, please also refer to the according setion in the OpenFold Readme file. In the examples, we assume anaconda3 is installed in the home directory. If this is not the case, the according paths have to be changed.

Inference from AlphaFold weights

python openfold/run_pretrained_openfold.py \
  fastas \
  data/pdb_mmcif/mmcif_files/ \
  --output_dir ./ \
  --model_device "cuda:0" \
  --uniref90_database_path data/uniref90/uniref90.fasta \
  --mgnify_database_path data/mgnify/mgy_clusters_2018_12.fa \
  --pdb70_database_path data/pdb70/pdb70 \
  --uniclust30_database_path data/uniclust30/uniclust30_2018_08/uniclust30_2018_08 \
  --bfd_database_path data/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt_ \
  --jackhmmer_binary_path ~/anaconda3/envs/openfold_venv/bin/jackhmmer \
  --hhblits_binary_path ~/anaconda3/envs/openfold_venv/bin/hhblits \
  --hhsearch_binary_path ~/anaconda3/envs/openfold_venv/bin/hhsearch \
  --kalign_binary_path ~/anaconda3/envs/openfold_venv/bin/kalign \
  --jax_param_path params/af_params/params_model_4.npz

Inference from OpenFold weights

python openfold/run_pretrained_openfold.py \
  fastas \
  data/pdb_mmcif/mmcif_files/ \
  --output_dir ./ \
  --model_device "cuda:0" \
  --uniref90_database_path data/uniref90/uniref90.fasta \
  --mgnify_database_path data/mgnify/mgy_clusters_2018_12.fa \
  --pdb70_database_path data/pdb70/pdb70 \
  --uniclust30_database_path data/uniclust30/uniclust30_2018_08/uniclust30_2018_08 \
  --bfd_database_path data/bfd/bfd_metaclust_clu_complete_id30_c90_final_seq.sorted_opt \
  --jackhmmer_binary_path ~/anaconda3/envs/openfold_venv/bin/jackhmmer \
  --hhblits_binary_path ~/anaconda3/envs/openfold_venv/bin/hhblits \
  --hhsearch_binary_path ~/anaconda3/envs/openfold_venv/bin/hhsearch \
  --kalign_binary_path ~/anaconda3/envs/openfold_venv/bin/kalign \
  --config_preset "model_1_ptm" \
  --openfold_checkpoint_path params/of_params/finetuning_ptm_2.pt