Multimeric bait protein with known structure

[t-fitzgerald]

Hello,

I’m currently using AlphaPulldown to screen for binders against a bait protein complex that is hetero-oligomeric. My bait is composed of 4 copies of protein A and 8 copies of protein B. I’d like to perform a pulldown mode screen while enforcing that the bait protein retains a structure close to the known hetero-oligomer configuration provided in a .cif file. I have looked over discussion #188 but I believe this may be a different use case.

• I ran generate_individual_features.py on the monomers from the bait (proteins A and B) as well as on the candidate proteins.
• Now, I'd like to run run_multimer_jobs.py in pulldown mode with --multimeric_mode=True. I specified a template directory containing the .cif file of the full hetero-oligomer and provided a description.csv that maps each monomer to its corresponding chain in the template. However, I do not have template files for the candidate proteins. Therefore, the candidate binders do not have corresponding template files or entries in description.txt. Is this an issue?
• See example command:
  python run_multimer_jobs.py \ --mode=pulldown \ --multimeric_mode=True \ --description_file=my/dir/alphapulldown/templates/description.csv \ --path_to_mmt=my/dir/alphapulldown/templates \ --msa_depth=1 \ --monomer_objects_dir=my/dir/alphapulldown/output/ \ --protein_lists=my/dir/alphapulldown/proteins.txt \ --output_path=my/dir/alphapulldown/multimer_output/ \ --data_dir=/my/dir/alphafold/db \ --num_cycle=1 \
• For the protein list, I formatted protein_list.txt as:
• "proteinA,4;proteinB,8;candidate" for each candidate. This should model the interaction of my heterooligomer bait with my candidate binders, correct?
• My main question is: How can I ensure that the provided .cif template is actively used during the structure prediction to constrain the bait protein’s conformation to the known structure? (I have set the --msa_depth flag to 1 in an effort to enforce this, but I don't think this is necessarily the best way to do this.)

Based on my reading of the documentation I feel like I'm missing something that would make this process more straightforward. Apologies if it is my own lack of understanding. Any guidance on the correct configuration or any additional flags I might need to set would be greatly appreciated! Thanks!!

[DimaMolod]

Hello @t-fitzgerald and thanks for you question. What you need to do is to create features for your bait protein chains using the provided .cif template and TrueMultimer mode. For the target proteins, you can use the standard pipeline for creating features. Then, when you are done and have all necessary features for both bait and target proteins, you can run_multimer_jobs.py with this additional flag --multimeric_template - please check the logs that the TrueMultimer mode is activated ("Running in TrueMultimer mode"). Please read more about how to run predictions with TrueMultimer in our documentation and feel free to ask further questions here is something is not clear.
BR,
Dima

[t-fitzgerald]

Hi @DimaMolod! And thanks for your prompt reply.

I have run the bait hetero-oligomer through the TrueMultimer mode of create_individual_features.py and run the standard pipeline for creating features for the target proteins.

Do the features created for the bait proteins from TrueMultimer mode need to go in to the same monomer_objects_dir as the features for the target proteins? How should I point the script to these custom features?

And would that mean that protein_list.txt needs to reflect the names of the .pkl files that were generated via TrueMultimer mode? That is,
ProteinA,4;ProteinB,8;target expands to ProteinA.TMPL.cif.A;ProteinA.TMPL.cif.B;ProteinA.TMPL.cif.C;...target etc?

This would make the new run_multimer_jobs.py call read as:

python run_multimer_jobs.py \
  --mode=pulldown \
  --multimeric_template \
  --msa_depth=1 \
  --monomer_objects_dir=/my/dir/alphapulldown/output/ \ # <- put all .pkl files here for bait and target?
  --protein_lists=/my/dir/alphapulldown/proteins.txt \ 
  --output_path=/my/dir/alphapulldown/multimer_output/ \
  --data_dir=/my/dir/alphafold/db_20231031 \
  --num_cycle=1 \
  --job_index=$SLURM_ARRAY_TASK_ID

I appreciate your support!

[DimaMolod]

Hi, yes, exactly, all features should be in the same directory, but you have to point the script to individual chains (read, .pkl files), because each .pkl file now contains individual template for each protein chain. That is also correct that you can do this with custom mode and protein_lists.txt. Please note, that using msa_depth=1 may lead to unstable predictions unless you increase number of cycles. As the rule of thumb I usually start with msa_depth=32 and num_cycles=3 to make sure AF follows the provided templates, before experimenting further. Please let me know if it works for you and feel free to reach me if you have any questions/doubts.
BR
Dima