mk_af_model.predict and localcolabfold

[vuhongai]

Dear authors, I found that the results of mk_af_model.predict and localcolabfold are quite different for prediction of 2-sequences complex. I wonder why there is such different? (I want to use mk_af_model.predict for screening because it's much faster than localcolabfold). Here's the parameters:

num_recycles = 3
use_multimer=True
model_names="model_1_multimer_v3"
initial_guess=False

Thank you for your help. Ai

[vuhongai]

Ah I think I got it, the limiting step is to get the MSA.

[Echavarria-Galindo]

Ah I think I got it, the limiting step is to get the MSA.

Hey mate, I'm also encountering the same problem as you and would like to know if you could give me some insights as to why the ColabDesign prediction (Through mk_af_model.predict) is so different to the LocalColabFold one. Is it because there is a simplified MSA version in ColabDesign?