Tutorial 5 Multimer predictions

Introduction

AlphaFold-Multimer is a computational tool for the prediction of protein-protein complex with known stochiometry.

LIT-AlphaFold was built on top of the architecture of AlphaPulldown which presents three different approaches to run calculations for large numbers of protein and protein multimers:

• Pulldown
• All vs all
• Custom
• Homo-oligomer

All of these methods are currently implemented in LIT-AlphaFold. Each prediction mode has its own specific use case and it should be selected based on the type of experiment you need to perform. The results from Tutorial 1 will be used as inputs.

Pulldown

NOTE: This method is reported only as an example, no input files are currently available for it.

In Pulldown mode combinations between different lists of proteins are generated. Each protein list count as a monomeric unit in the final multimer structure predicted by AlphaFold-Multimer. By passing two list of proteins all possible dimer structures between the member of list_1 and list_2 will be generated. By passing three list all possible trimers formed by member of list_1, list_2 and list_3 will be generated. Monomer modifications can be applied to both all proteins in a list, and in case group of proteins can be treated as a single unit. as an example given three input files list_1.in.yaml, list_2.in.yaml, list_3.in.yaml.

----
- CXCR4
----
- CXCR4
- CXCR4

- CXCL12

- G_alpha

The outputs will be the two complexes: CXCR4-CXCL12-G_alpha, and CXCR4-CXCR4-CXCL12-G_alpha

To run the calculations use the script as following:

python $LITAF/run_multimer_jobs.py \
mode=pulldown \
output_path=test_structure \
monomer_objects_dir=.[./tutorial_1_2/protein_features_mmseqs2] \
input_file=[list_1.in.yaml, list_2.in.yaml] \

All vs. All

NOTE: This method is reported only as an example, no input files are currently available for it.

The All vs All mode is used to generate all possible non-homo-oligomer dimers from a list of input proteins (with homo-oligomer in this case it is indicated that the same line in the input file will not be paired with itself). The single unit can be formed by more than a protein, as an example an input file like the one below:

----
- CXCR4
----
- CXCR4
- CXCR4
----
- CXCL12

Would generate the complexes CXCR4-CXCR4-CXCR4, CXCR4-CXCL12, CXCR4-CXCR4-CXCL12

To run the calculations use the script as following:

pypython $LITAF/run_multimer_jobs.py \
mode=all_vs_all\
output_path=test_structure \
monomer_objects_dir=[../tutorial_1_2/protein_features_mmseqs2] \
input_file=[protein.in.yaml] \

Custom

The custom mode has been showed before for monomer prediction, however its flexibility allows to use it also to predict more complex multimeric structures, without the need to predict all possible protein combinations. All units in a multimer are written as a list, with different multimer entries being separated by ----.

An example using CXCR4 and CXCL12 in multimer.in.yaml.

---
- CXCR4
- CXCL12

In this way it is possible to generate both dimers (CXCR4-CXCL12), and higher order multimers (CXCR4-CXCR4-CXCL12-CXCL12). This method also allows to modify only one monomer while leaving the other unchanged.

python $LITAF/run_multimer_jobs.py \
output_path=test_structure \
monomer_objects_dir=[../tutorial_1_2/protein_features_mmseqs2] \
input_file=multimer.in.yaml \

Homo-oligomer

The Homo-oligomer mode has been shown before to generate monomer structures, however it can also be used to generate dimers and higher order multimers. The input file for homo-oligomers generation is formed by the monomeric unit followed by a semicolon and the number of units in the homo-oligomer.

An exaple using CXCR4 in homooligomers.in

---
{monomer: CXCR4; n_units: 2}
---
{monomer: CXCR4; n_units: 3}
---
{monomer: CXCR4; n_units: 4}

This input file generates respectively: monomer, dimer, trimer of CXCR4.

python $LITAF/run_multimer_jobs.py \
mode=homo-oligomer \
output_path=monomer_structure \
monomer_objects_dir=../tutorial_1_2/protein_features_mmseqs2 \
input_file=homooligomers.in.yaml \

Conclusion

In this tutorial you have learned how to run multimer predictions using LIT-AF.