You are here

Native gives higher (positive) than lowest energy and clustered decoys

7 posts / 0 new
Last post
Native gives higher (positive) than lowest energy and clustered decoys


My native structure gives higher (positive) energy than lowest energy and clustered decoys (170 vs -77) with an output of 40000 decoys using abnitiorelax app.
The protein is an all beta-sheet and small (40 aa). From papers i`ve read and looking at the hydrogen bond terms at the rosetta energy function, i know it isn`t very good
at modeling this kind of proteins. I`m going to try it anyway. Any ideas why this happens? What am i doing wrong?

Thanks for the help,


Post Situation: 
Thu, 2011-06-02 07:19

It's known that structures from the PDB, when scored directly with the Rosetta energy function, can give ludicrously high energies. The reason is that Rosetta is sensitive to small positional differences. (e.g. if two atoms are slightly closer than they should be - we're talking fractions of an Angstrom - the VDW repulsive energy can shoot way up.)

It's highly recommended to either minimize the structure (or use the regular relax application) in the Rosetta energy function prior to comparing energies between native and predicted structures.

Thu, 2011-06-02 08:53

Using the idealize application or the relax application?

Thu, 2011-06-02 09:01

The idealize application is more to change bond distances and angles to their "ideal" settings.

The relax application ( would likely be your best bet. To keep the backbone from moving too much, you can add the flag "-constrain_relax_to_start_coords" to the command line.

By the way, you might also be interested in

Thu, 2011-06-02 15:43

Thank you very much for your time.


Fri, 2011-06-03 02:04

I´m assuming that relax uses scores12. For comparing the native score against decoys, generated with abinitio, the score weights should be the same.
Are they, in the default option, the same for relax and abinitiorelax?

thank you


Fri, 2011-06-03 08:16

According to the documentation (, the default for relax is score12 (that should be the default for most applications dealing with just proteins). If you're concerned, for most applications you can explicitly specify score12 weights with a "-score:weights score12_full" command line option.

Fri, 2011-06-03 11:56