# Cannot get a negative score after relax

7 posts / 0 new
Cannot get a negative score after relax
#1

Dear rosetta_users,
I am trying to relax a subunit of a actin filament(PDB ID: 3g37, chain O).
I have deleted all the modification and ligands, generated side-chains atoms by using MaxSprout, and add missing atoms by DeepView. Then I tried to use the relax function of rosetta before next step, with constraint. However, the total score of the structure is ~1200, which seems unreasonable. I also got OPT-H WARNING about chi angle, so I wonder if the high score is related with these warnings. What can I do to get a negative, at least much lower score of the structure?

core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 12 ASN ASN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 40 HIS HIS
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 41 GLN GLN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 59 GLN GLN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 77 ASN ASN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 86 HIS HIS
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 100 HIS HIS
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 114 ASN ASN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 120 GLN GLN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 127 ASN ASN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 245 GLN GLN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 262 GLN GLN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 274 HIS HIS
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 2 position: 296 ASN ASN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 313 GLN GLN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 353 GLN GLN
core.io.pdb.file_data: [ OPT-H WARNING ] heavyatom chi angle change: chidev= 180.000 chino= 3 position: 359 GLN GLN
core.scoring.rms_util: WARNING: In CA_rmsd, residue range 1 to 375 requested but only 374 protein CA atoms found.
core.scoring.rms_util: WARNING: In CA_rmsd, residue range 1 to 375 requested but only 374 protein CA atoms found.
protocols.relax.FastRelax: CMD: repeat 21376.6 0 0 0.44
core.pack.interaction_graph.interaction_graph_factory: Instantiating DensePDInteractionGraph
core.pack.pack_rotamers: built 17857 rotamers at 375 positions.
core.pack.pack_rotamers: IG: 67512436 bytes
core.scoring.rms_util: WARNING: In CA_rmsd, residue range 1 to 375 requested but only 374 protein CA atoms found.
core.scoring.rms_util: WARNING: In CA_rmsd, residue range 1 to 375 requested but only 374 protein CA atoms found.
protocols.relax.FastRelax: CMD: ramp_repack_min 210.535 0.110422 0.110422 0.0088
core.pack.interaction_graph.interaction_graph_factory: Instantiating DensePDInteractionGraph
.....

===================================================================================================================
Here is the flags I used for the relax function:
-in:file:s tmp_aa.pdb
-database /home/rosetta_2013wk40_bundle/main/database
-nstruct 2
-relax:constrain_relax_to_start_coords
-relax:coord_constrain_sidechains
-relax:ramp_constraints false
-ex1
-ex2
-use_input_sc
-flip_HNQ
-no_optH false
-ignore_zero_occupancy false

-out:output
===================================================================================================================
Here are the scores I got:
SCORE: total_score coordinate_constraint dslf_fa13 fa_atr fa_dun fa_elec fa_intra_rep fa_rep fa_sol hbond_bb_sc hbond_lr_bb hbond_sc hbond_sr_bb omega p_aa_pp pro_close rama ref description
SCORE: 1208.611 715.782 0.000 -1279.565 577.209 -53.508 5.461 502.412 740.565 -7.128 -26.767 -10.294 -57.000 93.660 -16.848 17.294 11.669 -4.332 tmp_aa_0001
SCORE: 1204.487 715.420 0.000 -1278.536 576.487 -53.405 5.431 500.339 739.925 -7.641 -26.184 -10.241 -56.865 92.320 -16.786 17.331 11.221 -4.332 tmp_aa_0002

The fa_dun(576.487) and fa_sol(739.925) also seem unreasonable.

Post Situation:
Tue, 2014-01-28 22:28
kxiao

Rosetta can (and in fact *has to*) place missing sidechains and hydrogen atoms itself. You don't need to use MaxSprout and DeepView to do that. (Deleting modifications and ligands outside of Rosetta is probably the recommended way to do it for maximum control, but that sometimes even that can be avoided with the -ignore_unrecognized_res option.) In likelihood that's what the OPT-H warnings are about - your MaxSprout/DeepView protocol is putting sidechains in conformations that Rosetta really doesn't like. That could have an effect on the downstream relax protocol - if you're constraining on those off conformations, you'll get poor scores, most likely in fa_dun. Try using the missing sidechain and hydrogen structures as input to relax, and see if that fixes things. (Or, alternatively, you can omit the "-relax:coord_constrain_sidechains" flag to avoid constraining the sidechains altogether, although that won't constrain those sidechains that are present.)

If it's not that, one issue may be the use of the always-on constraint protocol. This protocol is intended to minimize the atom movement of the structure. However, if you have major clashes or defects in your model, you won't be able to find a good structure without moving the atoms significantly. Try omitting the flag "-relax:ramp_constraints false". This tells the relax procedure to turn off the constraints at the end of the relax protocol, allowing you greater movement of the atom coordinates. This may allow Rosetta to find a good scoring model, at the expense of it being further from the starting structure. Be sure to examine the output models to see if you would consider the movements significant.

If that doesn't work, you may want to examine the output models in more detail, to see exactly where the bad energies are coming from. If you run the residue_energy_breakdown application on the relax output file, it will make a table which will report what each residue and each residue interaction is contributing to the score. You may have a small region of your complex which is resulting in the bad scores. (i.e. most of the model relaxes fine, but one region is "off", causing the total score to be bad.)

As a final note, with the always-on constraint protocol, the coordinate constraint value is contributing to the total score. So at least +700 of your +1200 is from the constraints. Even if you just rescored the same structure without constraints that would drop the total score down to +500 (which is still high).

Wed, 2014-01-29 12:49
rmoretti

Hi rmoretti, thank you for the information.

I tried to omit the "-relax:coord_constrain_sidechains" flag to avoid constraining the sidechains, and got lower scores:
SCORE: total_score coordinate_constraint dslf_fa13 fa_atr fa_dun fa_elec fa_intra_rep fa_rep fa_sol hbond_bb_sc hbond_lr_bb hbond_sc hbond_sr_bb omega p_aa_pp pro_close rama ref description
SCORE: 77.408 197.942 0.000 -1251.564 367.669 -98.025 3.619 247.041 703.726 -19.168 -33.817 -27.927 -66.933 67.448 -25.190 12.380 4.540 -4.332 tmp_aa_0002

The total score is negative if the coordinate constraint value is subtracted, but the fa_dun still seems high(~360).

I wonder if there are some rules of thumb for the rosetta scores, especially for each individual value. For instance, what total_score is suitable for a 374 aa protein? or how much score is acceptable for fa_dun value?

Sun, 2014-02-02 06:39
kxiao

The rule of thumb for score12 (Rosetta3.5 and before) was about -2 REU per residue for total score. (See figure 1 of Nivon et al http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0059004 for a graph.) The absolute magnitudes for talaris2013 total scores (the default scorefunction for the weekly releases) are a little less. I can't remember exactly, but it's about 2/3 that of score12, so around -1.6 REU per residue. (So "REU" really isn't a consistent or an absolute energy scale.) Unfortunately I don't have a good sense as to how all the sub-scores should relate to each other.

So you're still high, even when you don't constrain sidechains. My guess is that your actin filament structure probably comes from a low-resolution map, and as such has a much higher backbone positional error than other higher resolution structures do. You may want to reduce the stringency of the coordinate constraints (pass a value higher than 0.5 to -relax:coord_cst_stdev ), or allowing constraint ramping during relax (remove the -relax:ramp_constraints false flag). Those should allow greater backbone movement, which may fix your scoring issues.

Mon, 2014-02-03 15:54
rmoretti

Thank you very much for the update, I do really appreciate it!

Mon, 2014-02-10 06:19
kxiao

I remember someone said it is -2.5 REU/res for score12 without any reference and I always took that as the rule of thumb. That's in agreement with -1.6 REU if you say talaris2013 is ca. 2/3 of score12. Otherwise -2.0 * 2/3 = -1.3. In my own work I realize this value can often go below -2.0 but rarely reaches -2.5 REU/res.

From the PLoS ONE paper, it is unclear if -ex1 -ex2 had been used in unrestrainted relax procedure (they are specified for design).

Tue, 2014-04-08 06:59
attesor

D'oh - You're right, -2.0 * 2/3 = -1.3 But as I said, that's rough guess and slightly dependent on the protein, so seeing values of -2.0 is not too surprising.

Regarding the flags, if I remember correctly, -ex1 -ex2 were not applied for the relax even in the unconstrained state - others have done side-by-side comparisons of the (unconstrained) FastRelax procedure and seen that adding those flags don't actually change things all that much. (Likely because the minimization when repeatedly cycled with packing in the presence of lowered repulsive compensates for the coarseness of rotamer bins. This is something that's also seen with RTmin and the MinPacker - adding the ex flags doesn't help all that much when you have appropriate minimization involved. Note that design is a different situation, as you're making side-chain identity decisions prior to the minimization steps.) But again, I don't think that per residue scoring was necessarily one of the criteria by which they were explicitly compared, so perhaps there'd be a distinction there.

Tue, 2014-04-08 07:54
rmoretti