search

dina-lab3D / CombFold

Apache License 2.0
68 stars 12 forks source link

Folding homo-dimer filaments #14

Open Luke-ebbis opened 1 month ago

Luke-ebbis commented 1 month ago

Hey @ben-shor, right now I am attempting to use combfold to fold an homo octomer structure that is similar to human IMPDH1 and forms filamentous structures. Right now I want to make the complex for A. thaliana.

Input file.

Details

```json { "IMPDH1": { "name": "IMPDH1", "chain_names": [ "1", "2", "3", "4", "5", "6", "7", "8" ], "start_res": 1, "sequence": "MSTLEDGFPADKLF ... GVVGAVADKGSVLKLIPYTMHAVKQGFQDLGASSLQSAHGLLRSNILRLEARTGAAQVEGGVHGLVSYEKKSF" } } ```

Combfold has failed to assemble this octomer, even though the pairwise structures and the triplet combinations made with AFM all look like the structure produced by AF3--which did manage to fold the octomer.

Could you help me interpret the log file of the combfold programme?

Console log of the assembly process is attached.

output.log

In the output log, the programme notes that transformation files cannot be opened even though all files are present The programme also notes that it is skipping a transformation because of a 'missing interface':

... Skipping transformation, missing interface between IMPDH1 IMPDH1 ```

This error I don't understand because the AFM programme managed to come up with a homo-dimer and homo-trimer.

I checked your useage of Combfold in the publication for homo-decamers (5T11) and I notice that the input parameter end_res is missing for me. Could you tell me more about this parameter? Maybe it causes combfold to not detect the complex?

ben-shor commented 1 month ago

It seems CombFold is able to predict subcomplexes up to 4 subunits. (you can see it in the output " running iteration 5 prev kept results: 5" but " running iteration 6 prev kept results: 0") From looking at 7RFF, it seems that the complex is a "dimer" of two tetramers (ABCD and EFGH). That means there are 2 different types of dimeric interfaces, one to create the tetramer (for example AB) and one to joint the two tetramers (fr example AH). If you only predicted 2-3 subunits in your predictions, it is likely that the second interface was not present in your prediction so CombFold could not use it and so you will not get the correct structure. It is likely that if you will add a prediction of 5 copies of the subunit CombFold will work.

Hope this helps!