Before starting, please read the disclaimer at the end.

Installing AF2 locally

Dependencies and MSA

You can skip this section if you want to use our settings.

1. from https://colab.research.google.com/drive/1LVPSOf4L502F21RWBmYJJYYLDlOU2NTL?usp=sharing#scrollTo=a-COJivqdM8V copy the dependency cell into a file called "dependency_install.bsh"
2. Modify "dependency_install.bsh" with your settings. We use E_AMBER=False, USE_MSA=True, USE_TEMPLATES=False

Start Installing

Simply run

bash start.sh

This assumes you have conda installed.

Running Multiple Structures on the same GPU (Multiprocessing)

Running 1 structure at the time takes about 315MB of GPU. Using multiprocessing you could potentially run more structures on different workers.

python run_fold.py --workers 30 --num_models 1 --input_file /scratch/sequence-recovery-benchmark/monomers_af.json

run_fold.py accepts both .json or .fasta files

Credits

This work is hacked together by Rokas Petrenasand Leonardo Castorina from the ColabFold notebook from which dependency_install.bsh, msa2.bsh and run_fold.py are obtained. run_fold.py was modified to allow for multiprocessing and running multiple structures automatically.

As with ColabFold we would like to credit and thank:

• RoseTTAFold and AlphaFold team for doing an excellent job open sourcing the software.
• Also credit to David Koes for his awesome py3Dmol plugin, without whom these notebooks would be quite boring!
• A colab by Sergey Ovchinnikov (@sokrypton), Milot Mirdita (@milot_mirdita) and Martin Steinegger (@thesteinegger).

Disclaimer

As per https://twitter.com/thesteinegger/status/1420055602970075138 be mindful of how you use this repository. The API is currently supported by only one server handling multiple thousands of requests per day. Refrain from using this tool until they have improved the API (we will keep this up to date!)