VIBFold: an adapted AlphaFold script

Important: VIBFold is still under construction and being tested.

The default AlphaFold script that comes with the installation works fine, but it leaves room for improvements in terms of user-friendliness. Therefore, we adapted existing scripts to obtain an updated version with extra customization options, more visualizations and further automation.

It is important to note that this script is strongly based on the ColabFold repository (Mirdita et al, 2021), and both code fragments and the visualization image strategy have been copied from there, as well as from the official AlphaFold repository.

VIBFold implements following improvements to the existing AlphaFold script:

• Extra customizability options. Currently available:
  • MSA search: choice between the default AlphaFold search, and use of the MMSeqs2 web server (not supported yet)
  • Enable or disable the use of templates
  • Allow more recycling steps (= re-iterations of the prediction model using the previous prediction – default=3)
  • Enable relaxation for all predicted models, only for the best, or none. Useful to deal with time constraints for large proteins.
• For monomer prediction:
  • Multiple jobs can be launched at once, using a single FASTA file.
• For multimer prediction:
  • As in the default script, all entries in the FASTA are modeled into one protein complex
  • Additionally, all possible permutations of the FASTA entries will be run automatically through AlphaFold-Multimer. For proteins A and B, modeling A+B can give different outcomes than B+A.
• Any directory can be specified for outputs
• Visualizations are added (cfr. ColabFold):
  • The MSA coverage and sequence identity
  • pLDDT graphs for the predicted structures
  • PAE graphs for the predicted structures
• Large files (such as the MSA and .pkl files are automatically removed for the sake of storage efficiency)