running Rosetta locally using commands generated by the Robetta web server

[Anonymous]

Dear Rosetta users:

I ran a full structure prediction on the Robetta web server and obtained useful results. Now I would like to run Rosetta locally (because I want to control the fragment libraries).

I installed Rosetta3.5 and did a test run, but the locally-generated structures are quite different than those that I obtained from the Robetta web server (the Robetta results are much better). I presume that this is because I have not configured the local Rosetta run very well.

What I would really like to do is to obtain a copy of the commands that were sent to Rosetta from the Robetta web server so that I can repeat the structure determination locally and get the same results that I got from the Robetta web server. Is that possible?

Thank you,
Chris.

[Anonymous]

First off, if you used Robetta recently, the results will likely not be comparable to a local run with Rosetta3.5, as Robetta is currently using a version of Rosetta akin to weekly releases. There are substantial differences between 3.5 and the weekly releases, most notably, the main fullatom scorefunction was changed from score12 (in Rosetta3.5) to talaris2013 (in the weeklies). I’d expect that change alone to have a significant difference in result.

Additionally, it’s highly likely that Robetta is simply using much more sampling than you did locally. Generally speaking, you need rather extensive sampling in order to get decent results with ab initio structure prediction.

I unfortunately do not have access to the commands which Robetta is running. I’ve forwarded your request to someone who does, though, and hopefully he’ll be able to provide more information.

[bweitzner]

Can you provide me the job id? Email it to dekim at u dot washington.edu

If it still exists on our filesystem, I can send you the log files which have most command lines etc… and I’ll know what protocol you are using, i.e. ab initio or comparative modeling.

[Anonymous]

Thank you both, rmoretti and dktest. Indeed, my local sampling was very limited (job took a few minutes to run on a single core). I have lots of additional resources that I can throw at this, I’m just hoping to know exactly what Robetta did as a quick way to learn how to do an ab-initio Rosetta run “correctly”. I have emailed dktest with job id and look forward to hearing back.

Thanks again,
Chris.

[Author]

Posts