HPC: Boltz-1

Apart from AlphaFold2 (and OpenFold), another protein structure predictor installed on the HPC is Boltz-1.

Boltz-1 can be run by either providing your own MSA, or by letting the program use an online web service to generate the MSA. In this exercise, we make use of the web server option.

Using online web server to compute the MSA, then run Boltz-1

Boltz-1 accepts a .yaml file (with protein sequences and possibly other biomolecules) in input. You can find an example of a .yaml file here. Then, you need the job script that sets the option to run the MSA search online. You can find an example job script here. Download the files, set the correct directories in the job script, and submit as with previous job scripts.

Results

The first run, it has to download the model parameters (by default stored in $VSC_DATA). After that, prediction will start for a couple of minutes for a small protein, after which you should already have the results.

The results contain following subdirectories:

  • lightning_logs: logging files for information
  • msa: the processed MSA files
  • predictions:
    • the predicted structure (.cif file)
    • the model confidence (.json file) where you can find ptm, iptm and other scores
    • plddt_xxx.npz (numpy file) where the individual pLDDT values are stored
    • pae_xxx.npz (numpy file) where the individual PAE values are stored
  • processed: intermediate files

Previous Topic
Back to Lesson
Next Topic