You are here

Determine the weights of score functions

4 posts / 0 new
Last post
Determine the weights of score functions

Hi all,

Does anybody know how to calculate the weights of score functions? Generate a set of decoys and do a regression to find a good correlation between energy and rmsd?

Could somebody give me some suggestion? Thanks very much!


Thu, 2010-09-02 07:20

Yes - use the OptE protocol. It's not published or released. Someone's supposed to be writing a paper on it, but it's not yet close to publication. Recent or soon-to-be-published publications from the Baker lab (look for Kellogg in the author list) touch on this as well.

If you just want to know what the weights ARE, look in (database)/scoring/weights. You can change these files as you please to test different weights.

Fri, 2010-09-03 07:01

Thanks smlewis very much.

I'm using rosetta3.1 to dock a peptide to a protein. I found that my protocol could sample the near-native conformation but it's energy wasn't the lowest. Score function used in my protocol was standard.wts patched by score12.patch.

I had tried several other score functions but the results were the same. So I think I need to change the weights of score functions to find a good correlation between decoy energy and rmsd. I've sent an e-mail to Rosetta-commons two weeks ago but got no reply till now. Could you tell me the protocol to calculate the weights?


Sun, 2010-09-05 07:34

Well, like I said, the protocol is known as OptE, and it's not published yet. The Top7 paper includes a very vague description of how score12 was originally weighted, which eventually grew into today's OptE. There's a paper in the works.

Short of that protocol being published, you can always try manually exploring weight-space by modifying the weights files. Try a weight set, see if it improves, try the next, etc. It's very laborious, but it's been done in the past...

Wed, 2010-09-08 07:30