You are here

Choosing weight file for interface design?

3 posts / 0 new
Last post
Choosing weight file for interface design?

Hi, I am now trying to do an interface design of a nanobody docked to the target (through its CDRs, by patchdock).

I have some doubt about choosing the best weight file for scoring of the designs. May I have some suggestions from you?

Actually, I have read from here and I have already learned that:
1) Choose score weighting by finding what was used for the most similar application in published literatures.
2) Use the weighting of ref2015, which serves as a general score function.
3) I can try playing around the weighting myself but the testing would be slow and tedious.

Then I have a few questions:
1) For the existing weight files under ...rosetta_bin_linux_2018.09.60072_bundle/main/database/scoring/weights/ , where could I find the manual (including the scoring rationale, who wrote it, primary citation etc.) of each weight files? For example, interface.wts, no description is present in the header of the file.

2) If I want to test the performance of existing weight files (e.g. interface.wts vs ref2015), do I have to use a benchmarking test set to test the performance of scoring interface closest to the reality (e.g. with experimental Kd)?

3) Is there an efficient method of googling .xml files? Do we have a library of published .xml files so we could look up previous methods efficiently?



Post Situation: 
Tue, 2019-05-07 00:47

1) Unfortunatly for a bunch of the weights files in the database, there isn't a good history of where they come from. Especially for the older weights, we didn't have a good policy for annotating them as to their source, and there isn't necessarily a reference which discusses how they were derived. Often the best you can do is consulting the oral history of people who have been in the Rosetta community for a while.

2) Yup, in order to test performance, you'd either need to find someone who has done the benchmarking for you, or figure out your own benchmark set to compare the two. How exactly you determine what's in that set and how to evaluate performance is somewhat of a black art of benchmark construction.

3) I'm not sure if there's a good way of doing such searching. We do distribute example XML files for various protocols in the Rosetta/demos directory and the Rosetta/main/rosetta_scripts_scripts directory. Depending on what you're doing, you may or may not find a recent XML script in one or both of those locations.


Regarding interface.wts and ref2015.wts, I'd probably recommend just sticking with ref2015.wts for design purporses, unless you have a compelling reason to use the older interface.wts file.  If I recall correctly, the interace.wts file was created for protein-protein docking back in the score12 era. There's several issues here. The first is that it's been optimized for fixed sequence docking, and may not contain all the terms at the weights needed to properly account for sequence change differences. The seconds is that it's been optimized for the score12 environment. There's a number of underlying functional form differences and parameter changes not embodied in the weights file which are different between score12 and ref2015, and I wouldn't necessarily expect a weights file optimized in the score12 environment to also be optimized in the ref2015 environment. (Though you can go back to the score12 environment with the -restore_pre_talaris_2013_behavior flag.) Finally, the reason for the parameter differences is that we've spent a bunch of time optimizing the scorefunction. A priori, I would expect that the ref2015 environment would perform better than something based on score12. -- As far as I'm aware, the people doing protein design are doing so with ref2015 or one of the newer experimental "beta" scorefunctions.

Thu, 2019-05-09 15:22

rmoretti, thanks for the suggestion! I think I will just stick with ref2015.wts unless there's something wrong.

Thank you!

Tue, 2019-05-21 22:33