This is adapted from this repo

Setup python environment

• conda create -n alphafold python=3.8 (rename alphafold as needed)
• conda activate alphafold
• conda install -y -c conda-forge openmm==7.5.1 cudnn==8.2.1.32 cudatoolkit==11.1.1 pdbfixer==1.7
• conda install -y -c nvidia cuda-nvcc==11.3.58 (this is needed as jax requires a functional ptxas binary)
• conda install -y -c bioconda hmmer==3.3.2 hhsuite==3.3.0 kalign2==2.04
• pip install -r requirements.txt
• pip install jaxlib==0.1.69+cuda111 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

The environment is almost ready, but a file needs to be patched. Assuming your base conda directory is in ~/miniconda3, run the following:

• cd ~/miniconda3/envs/alphafold/lib/python3.8/site-packages
• wget https://raw.githubusercontent.com/epfl-lts2/alphafold-setup/master/openmm.patch
• patch -p0 < openmm.patch

Download files

• mdkir /mnt/scratch.../workdir (adapt to your needs)
• cd /mnt/scratch.../workdir
• wget -q https://github.com/deepmind/alphafold/archive/refs/tags/v2.2.3.tar.gz && tar -xzf v2.2.3.tar.gz
• cd alphafold-2.2.3
• wget -q -P ./alphafold/common/ https://git.scicore.unibas.ch/schwede/openstructure/-/raw/7102c63615b64735c4941278d92b554ec94415f8/modules/mol/alg/src/stereo_chemical_props.txt
• wget -q https://raw.githubusercontent.com/epfl-lts2/alphafold-setup/master/run_alphafold.sh
• chmod +x run_alphafold.sh

Run alphafold

• Create a fasta query, e.g.

>pdb|3h7p|A
MQIFVKTLTGKTITLEVEPSDTIENVKAKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQRESTLHLVLRLRGG

• Run alphafold (the necessary databases have been already downladed and processed in /mnt/scratch/alphafold):

./run_alphafold.sh -d /mnt/scratch/alphafold -o test2 -c reduced_dbs -f [fasta input] -t 2020-05-14 -a [gpu id] -n [number of cpus]

e.g.:

./run_alphafold.sh -d /mnt/scratch/alphafold -o test2 -c reduced_dbs -f ./ubi.fasta -t 2020-05-14 -a 1 -n 16