After successfully predicting a monomer structure using AlphaFold, only slight changes to the setup are required for predicting a protein complex with AlphaFold-Multimer.
Information can be found at https://elearning.vib.be/courses/alphafold/lessons/alphafold-on-the-hpc/topic/extra-alphafold-multimer/.
An example complex, which will be discussed in further exercises, is the SARS-CoV-2 RBD bound by the single-domain antibody SARS VHH-E (7KN5 entry in PDB). The FASTA file can be downloaded here (note that we will only consider the RBD and VHH-E nanobody in this exercise!), but feel free to bring your own protein complex. The total time required for predicting the example on the joltik cluster is greater than before, as an MSA is computed for each input sequence. It will require around 2 hours.
Apart from supplying a FASTA file with multiple entries, you will need to cover following steps:
--num_runs_per_model=1 flag to only run each of the five AF-multimer models once (for the sake of this exercise — for optimal predictions it is advised to have multiple prediction attempts) (see also the example job script here)
And finally, submit the job script as before.
If you want to set up a prediction for your own FASTA sequences of interest, this would be a good time to put them on the HPC. The following exercises are focused on analysis, and no longer require GPU computation time on the HPC.