search

JeffSHF / ColabDock

Code for ColabDock paper
Other
114 stars 12 forks source link

Distortion of input structures #14

Closed s-kyungyong closed 9 months ago

s-kyungyong commented 11 months ago

Hi!

I am running the software locally. The best predictions seem to have some distortions in the ligand structure. For instance, the alpha helix in the original structure (colored in cyan) turned into smooshed coils possibly to make the specified contacts with the receptor possible. Is there a way to prevent this behavior?

image

Thank you!

JeffSHF commented 11 months ago

Hi, there. Are the restraints provided phisically possible to be satisfied? It will be helpful to give some details about the restraints. I encountered this situation before. Like you said, ColabDock may give unreasonable structures to satisfy the provided restraints.

s-kyungyong commented 11 months ago

The positional restraints make sense, but the receptor is quite rigid and the local environment where the binding happens is very compact. Every restrains were given to rest_1v1, and res_thres was set to 8.0. As I increased res_thres values, the distorted structures seemed to go away. But I wondered if there would be other tricks.

JeffSHF commented 11 months ago

You mean the binding will lead to large conformational change? This may be a difficult case. Maybe you can try to reduce the weight of dgram during the generation stage (https://github.com/JeffSHF/ColabDock/blob/3b62a74502244bb9953922b4c718d0e376c94234/colabdock/docking.py#L41C9-L41C17). By default, it is set to 1.5.

s-kyungyong commented 11 months ago

No, both input structures are sturdy and won't change much upon binding. I will play a bit more with some parameters!