Non Docker Alphafold

This is the modified version of Alphafold 2 that does not require docker.

In this pipeline, the program accepts an alignment file in a3m format. (Note: This program does not generate MSA files). It helps to evaluate the given MSA file using alphafold prediction results.

Install conda using miniconda if not installed already

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

Create conda environment

conda create --name <env_name> python==3.8
conda update -n base conda

Activate conda environment and install dependencies

conda activate <env_name>

conda install -y -c conda-forge openmm==7.5.1 cudnn==8.2.1.32 pdbfixer==1.7
conda install -y -c bioconda hmmer==3.3.2 hhsuite==3.3.0 kalign2==2.04

Install other alphafold dependencies using pip

pip install absl-py==0.13.0 biopython==1.79 chex==0.0.7 dm-haiku==0.0.4 dm-tree==0.1.6 immutabledict==2.0.0 jax==0.2.14 ml-collections==0.1.0 numpy==1.19.5 scipy==1.7.0 tensorflow==2.5.0

Change jaxlib version for alphafold

pip install --upgrade jax jaxlib==0.1.69+cuda101 -f https://storage.googleapis.com/jax-releases/jax_releases.html

Download Alphafold parameters

wget https://storage.googleapis.com/alphafold/alphafold_params_2021-07-14.tar -P <path_to_params_dir>

cd <path_to_params_dir>
mkdir params/

tar --extract --verbose --file=<path_to_params_dir>/alphafold_params_2021-07-14.tar -C params/

rm <path_to_params_dir>/alphafold_params_2021-07-14.tar

OpenMM Patch

Note: <path_to_alphafold> should the directory path where the alphafold directory is located

cd ~/miniconda3/envs/<env_name>/lib/python3.8/site-packages/
patch -p0 < <path_to_alphafold>/docker/openmm.patch

How to run?

bash run_alphafold.sh -d <path_to_params_dir> -o <output_dir> -m model_1,model_2,model_3,model_4,model_5 -f <path_to_fasta> -s <path_to_a3m_file> -t 2019-05-14

• -m: at least one model name must be provided
• -t: template date (refer alphafold github repository for more details)

Note

This codebase runs on CUDA 10.1. This was tried and tested in Ubuntu 18.04.4 LTS and the hardware specifications of the server are as follow:

1. Dual 4215R 3.2GHz CPUs
2. 128 GB RAM
3. NVIDIA Quadro RTX 6000 GUPs each with 24GB memory

Acknowledgement

To create this code base, the actual alphafold repository [https://github.com/deepmind/alphafold] was used as well as another repository from https://github.com/kalininalab/alphafold_non_docker was referenced.