fixbb algorithm reference?

5 posts / 0 new
fixbb algorithm reference?
#1

Hi,

I'm interested in understanding how the fixbb application works. It mentions in the documentation page:

a stochastic simulated annealing approach is used. However, I do not see a description of this approach in the references on that page.

I did find a reference to a Metropolis Monte Carol procedure here:

As an example, if I design an entirely new sequence of amino acids (specified by a resfile) for a 10 aa length protein, what happens?

I assume:

1. Intitial rotamers for all residues are randomly picked

2. Energy is calculated

3. A random rotamer substitution is made to a randomly picked residue

if (new energy < old energy {

accept new rotamer;

} else {

accept rotamer based on  some distribution with a temperature parameter (annealing step)

}

4. Repeat step 3 until some satisfactory point

Is that close?

Thanks

Category:
Post Situation:
Wed, 2015-09-23 10:20
tylerborrman

If you want information on how Rosetta handles this specifically, start here: https://www.rosettacommons.org/docs/latest/getting_started/Rosetta-canon at the Top7 paper.  There are several papers published after that (all by Andrew Leaver-Fay, with Brian Kuhlman and usually Snoeyink in the author list) that discuss how the packer works in greater detail.  Your description is accurate in the broad scale.

If you want to learn about Metropolis Monte Carlo in general, it's 50 years old, I assume wikipedia has the original references.

Wed, 2015-09-23 10:24
smlewis

However, the Top7 paper does not describe the annealing algorithm in the text or supplement. It does reference the PNAS paper I mentioned earlier in the thread, which references the J Mol Biol paper:

This paper describes a Monte Carlo approach to identify minimum energy amino acid side-chain conformations.

Is this 15 year old reference still the methodology behind the fixbb app?

Again, looking at the later papers referenced here,

I can not find any overview of what the fixbb app does when given a pdb and a resfile of mutations?

Thu, 2015-09-24 08:30
tylerborrman

The basic algorithm for fixbb hasn't changed significantly since it was first added to Rosetta, so a 15-year-old reference would be the contemporary one.

That's not to say there hasn't been any work on improving things. The Andrew Leaver-Fay papers would be the ones to look at, particularly: http://www.ncbi.nlm.nih.gov/pubmed/15759610 and http://dx.doi.org/10.1371/journal.pone.0020937 and http://www.ncbi.nlm.nih.gov/pubmed/21187238 That last one is a good overview of Rosetta in general. The fixbb application is basically just the "packer" which is discussed (briefly) there.

Also, looking at the references, http://dx.doi.org/10.1073/pnas.97.19.10383 looks to be one which is referenced as the original reference for the standard packer.

On what the fixbb application does, it more-or-less just loads the PDB into the Rosetta-internal representation of a structure (the Pose), then loads the resfile into a "PackerTask" which represents which amino acids and which rotamers are allowed at each position, then calls the packer to optimize the combination of amino acids/rotamers across the entire protein on the fixed backbone structure.

The treatment of design in these cases is entirely equivalent to the treatment of rotamer repacking, but instead of choosing between two rotamers of the same sidechains, it chooses between two rotamers which happen to be from different amino acids. The method by which is selects the rotamers is the Metropolis Monte Carlo Simulated Annealing approach, guided by the standard Rosetta energy function.

Fri, 2015-09-25 12:15
rmoretti

Excellent! Thanks for your help. The link to the PNAS paper you referred to in the thread appears broken. I believe this is the paper:

Mon, 2015-09-28 09:29
tylerborrman