You are here

dun10

6 posts / 0 new
Last post
dun10
#1

Hi,

Just curious if there is or there will be a flag to use Roland's 2010 Rotamer library within Rosetta. We have been replacing the 2008 binaries with data using the 2010 library and then using the 2008 flag, which works well enough. I'm still learning C++, or I would try and implement it myself.

Also, if it does already work within Rosetta, where would I find it within the code, and which level of smoothing was used? (optim, 5%, 20%)? We have been using optim (with the neighbor-dependent ramachandran flag) across most of Rosetta's apps and it seems to work well.

-Jared

Post Situation: 
Thu, 2012-01-19 12:16
jadolfbr

Nobody's put a copy of the 2010 library in trunk yet (do you have trunk access, right?). So, um...yeah. Swapping out the 2008 for the 2010 library and telling the code to use 2008 will work in most respects. The data will be fine, but the scorefunction was weighted against 2008 not 2010, so it will be suboptimal. (Actually, it may have been weighted against pre-2008, so it's equally suboptimal in both cases).

If you are interested in getting the 2010 library put in, I'll put you in contact with Andrew, who knows the rotamer library code best.

I haven't a clue about smoothing.

Thu, 2012-01-19 14:19
smlewis

Actually, Andrew already put the dun10 library in trunk. You can enable it with the -corrections:score:dun10 flag. It doesn't appear to be in the 3.3 release, although it likely will be included in the 3.4 release.

I believe you're correct that no one has yet optimized the scorefunction for use with the dun10 library. Yifan Song put in a bit of effort to optimize against the dun08 library, but it was soon discovered that dun08 had some issues (which in part lead to Dunbrack et al. putting out the dun10 library), and the effort has not yet been replicated for dun10. Use of the dun08 library with Rosetta is *highly* discouraged. I assume any dun08-specific optimization should have been backed out/turned off, so everything in Rosetta should be optimized around the (default) dun02 rotamer library.

Given that the dun10 and dun08 libraries are basically the same format, for people who don't have access to/don't want to use trunk, there is probably no reason not to replace the dun08 libraries with the dun10 libraries, as long as you remember you've done so when you write up the paper or discuss your procedure with others.

Fri, 2012-01-20 11:23
rmoretti

Knowing it was there made me look harder. "ExtendedOpt1-5" seems a suboptimal name for the library's directory...

Fri, 2012-01-20 11:40
smlewis

Thanks for the info.
The 2010 library looks like it works better then the 2002 even without optimization, but only the Fast and Classic relax was benchmarked with it and published. With an optimized scorefunction, we will probably be benchmarking it further in loop modeling and design later down the road...

Sat, 2012-01-21 15:04
jadolfbr

Thanks! I Should have looked there first...

Fri, 2012-01-20 08:25
jadolfbr