Open Huilin-Li opened 5 months ago
This is entirely dependent on your problem statement. You often have a pre-existing notion of where the binder is supposed to go. For example, if you want to make a protein which inhibits binding of a third protein to your target, then you probably want your designed binder to also bind the same interface that the third protein does. Even if you aren't inhibiting a known binding site, then you likely have a sense of where on the target protein you want to bind -- in a convenient pocket, away from the active site, on the membrane distal surface, etc. The hotspot approach is designed around that use case, where you want to pre-specify where on the target protein the designed protein should bind.
If you don't have any inclination of where the designed protein should go, you should just be able to omit the hotspot designations. I believe that without the hotspot designations (but still with the target protein template provided), RFDiffusion should still attempt to create a new protein which works in the context of the provided target protein. Where it binds won't be specified, but if you do a design run and don't like where RFDiffusion is putting the binders, you can always go back and specify hotspots that define a putative binding site. (Feel free to try multiple different runs with different hotspots each.)
This is entirely dependent on your problem statement. You often have a pre-existing notion of where the binder is supposed to go. For example, if you want to make a protein which inhibits binding of a third protein to your target, then you probably want your designed binder to also bind the same interface that the third protein does. Even if you aren't inhibiting a known binding site, then you likely have a sense of where on the target protein you want to bind -- in a convenient pocket, away from the active site, on the membrane distal surface, etc. The hotspot approach is designed around that use case, where you want to pre-specify where on the target protein the designed protein should bind.
If you don't have any inclination of where the designed protein should go, you should just be able to omit the hotspot designations. I believe that without the hotspot designations (but still with the target protein template provided), RFDiffusion should still attempt to create a new protein which works in the context of the provided target protein. Where it binds won't be specified, but if you do a design run and don't like where RFDiffusion is putting the binders, you can always go back and specify hotspots that define a putative binding site. (Feel free to try multiple different runs with different hotspots each.)
Thanks for your clear explanation! It is cool! In my past experiences, I picked top 5 very hydrophobic AAs who are most closest to the center of the target protein. I got one designed binder with pae < 10 in 100 designs. I will try your suggestions, too. Thanks agian!
I have a stupid question to ask, about this sentence "In the paper we define a hotspot as a residue on the target protein which is within 10A Cbeta distance of the binder".
I am working on selecting the specific AA position. Then, I find this sentence. If the "10A Cbeta distance of the binder" is the distance between target to binder, I am confused with it. Because we are going to design the binder, which means we don't know where the binder is, so how to calculate the distance between the target and the binder?