The background image is designed by starline / Freepik
www.hegelab.org
Our manuscript on AlphaFold2 perfomance and capabilities is online:
T. Hegedus, M. Geisler, G. Lukacs, B. Farkas:
Ins and outs of AlphaFold2 transmembrane protein structure predictions
Cellular and Molecular Life Sciences, DOI: 10.1007/s00018-021-04112-1
Our AlphaFold-Multimer fork
  • We have modified the AlphaFold2 code to avoid memory issues you may encounter when running original AlphaFold locally. This is expected to be included in the next AF2 release (see: pull request #296).
  • Please see details and download at github/hegelab/alphafold.
BetaFold-Multimer, a fork of AlphaFold2-Multimer
    We have modified the AlphaFold2 code to:
  • Avoid memory issues (see above).
  • Run different parts separately. This allows database searches on a CPU node and running models on a GPU node.
  • Please see details and download at github/hegelab/betafold.
Our AlphaFold2 fork (initial release)
  • This is a fork with minimal changes to the original AF2 code.
  • We altered a few parts to decrease the memory burden of sequence features generation.
  • This make the following patch obsolete.
  • You can access the fork at github/hegelab/alphafold
  • Please see details in our pull request
BetaFold, a fork of AlphaFold2 (initial release)
    We have modified the AlphaFold2 code to:
  • Run different parts separatly. This allows database searches on a CPU node and running models on a GPU node.
  • Predic complexes. Without a linker.
  • These can be done via configuration files.
  • Please see details and download at github/hegelab/betafold
Patch for "ValueError: Cannot create a tensor proto whose content is larger than 2GB."
This is already included in the current AlphaFold-Multimer release.
It seems that this is caused by two lines missing from the pipeline.py script: 
  uniref90_msa = uniref90_msa[:self.uniref_max_hits]  # hege
  uniref90_deletion_matrix = uniref90_deletion_matrix[:self.uniref_max_hits] # hege
Download the pipeline.patch file to your alphafold folder, and apply it as: 
  patch alphafold/data/pipeline.py pipeline.patch
Modified Alphafold2 scripts for skipping structural templates in prediction runs:
By default AlphaFold allows excluding structural templates based on date (--max_template_date). We introduced the option to skip specific PDB entries (--skip=pdbid1,pdbid2,pdbidx or --skip=skippdbfile). Please download, read the README.md file, and cite our paper (thanks).
hegelab-alphafold-skip-pdb.tgz