Succes rate

[alejandromontesa-unizar]

Is there any proof that using hotsposts is better than unconditional design. I assume that if hotsposts are well selected it can be better, but when you have no clear idea of the mechanism of a protein, do you think is better to run unconditional design. Looking forward to read your opinions.

[roccomoretti]

My understanding is that hotspots are used primarily to direct generation, not to improve it. That is, when you do free design, RFdiffusion is permitted to create designs wherever it likes. Some may have interactions with the regions you're interested in, but some might only contact it very lightly, or might be associated with the other side of the partner. In contrast, when designing with hotspots, RFdiffusion will focus its attention on designs which have contacts in the region of interest, so you'll get a larger fraction of designs that are located in the region of interest.

But other than that, I'm not aware of any quality difference in the designs. That is, if you filter by "does this interact with the residues I'm interested in", a hotspot design will give you a larger fraction of designs which pass the filter than a free design would, but there isn't necessarily any difference in quality in the designs which do pass the filter.

So if you don't have specific residues you're interested in interacting with for binder design, there probably isn't a reason to use hotspots.

[alejandromontesa-unizar]

Thanks for you answer @roccomoretti I agree with you. And what about binder legnth, any thoughts or opinions on that?

[roccomoretti]

I'm not aware of papers which have specifically looked at the correlation between generated protein length and binder design success, but that doesn't mean there aren't any out there. That said, my impression is that binder length is typically driven by experimental demands and other downstream considerations (e.g. how big a protein are you able/willing to synthesize).