Hello,
I am new in Rosetta. I would like to compare stability of the wild-type and mutated proteins using DDG_monomer. Since my protein is quite big I use the low resolution protocol for now. The results that I obtained look weird to me. Could you please tell me if they are supposed to be like this or I do something wrong?
I obtained the following output files: ddg_predictions.out (it looks fine to me), wt_S360D.out (all structures and corresponding scores are either the same or have negligible differences), mut_S360D.out (all structures and corresponding scores are either the same or have negligible differences), mutant_trajS360D (empty!), wt_traj (empty!). Please see my flag file and grepped scrores below.
According the application documentation, I expect conformational sampling of sidechains within 9A of the mutated residue for the both wild-type and mutant proteins. But the structures I obtained look like I applied some strong constrains to all atoms. There is no any sampling of rotamer conformations.
Thank you in advance!
Maria.
@flag_ddg_low
-in:file:s input_files/min_cst_0.5.5flz_C_0001.pdb
-ddg::mut_file input_files/mut_file.dat
-ddg:weight_file soft_rep_design
-database /gs/project/ear-065-aa/rosetta_src_2017.45.59812_bundle_12nodes/main/database/
-fa_max_dis 9.0
-ddg::iterations 50
-ddg::dump_pdbs true
#-ignore_unrecognized_res
-ddg::local_opt_only true
-ddg::suppress_checkpointing true
-in::file::fullatom
-ddg::mean true
-ddg::min false
-mute all
-ddg::output_silent true
-out:level 400
wt_S360D.out SCORES
SCORE: score fa_atr fa_rep ...
SCORE: -686.234 -3344.916 279.393 ...
SCORE: -686.234 -3344.916 279.393 ...
SCORE: -686.234 -3344.916 279.393 ...
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -685.691 -3344.808 278.848
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
SCORE: -686.234 -3344.916 279.393
Which row in the Kellogg paper are you trying to emulate? From the flags I'm guessing it's row 3 (repack everything within 8 Ang with the soft-rep scorefunction, with no minimization of either sidechain or backbone), but your flags are slightly off from those listed in the Kellogg paper SupMat.
Specifically, you don't have the `-ddg::min_cst false` option set. (The `-ddg:min` option tells ddg_monomer to take the minimum value from multiple -ddg::iterations -- it doesn't say anything about structural minimization.) The default for the -ddg::min_cst is true, so your settings involve minimization. Since -ddg:sc_min_only defaults to true, though, it's sidechain only minimization.
Also the -fa_max_dis setting is not setting the radius for repacking. It's setting the cutoff radius for the attractive portion of the van der Waals potential. To set the repacking radius, you need the -ddg::opt_radius option with the -ddg::local_opt_only option set to true. (-ddg::opt_radius defaults to 8.0, though, so you're not terribly off.)
I don't think that's related to what you're seeing with consistency, though. Instead, I think that's just a symptom of the fact that for fixed sequence repacking on fixed backbones, the packer is reasonably convergent. It depends a bit on what the structure and sequence around the mutating residue is, but repacking in an 8 Ang shell around a single residue is a relatively small optimizational space, as these things go, particularly if the mutation is not that disruptive and if the input PDB was well optimized. It does look like you get some variation in the output (the one -685.691 structure), but for the most part Rosetta is able to find the better -686.234 structure. It's not that it isn't sampling the other structures - it just samples them internally along with the -686.234 structure. It then recognizes the -686.234 is better than the other structures and discards them. The only reason you'd get structures other than the -686.234 one is those rare cases where the stochastic search isn't able to pick up the -686.234 one and has to settle for a worse one.
It's true that having -ddg::iterations set to 50 is probably overkill in your case. But when you get to larger sampling (full protein optimization especially with minimization and backbone ensemble generation) it then becomes helpful to do a wider sampling. I'm guessing Kellogg et al just settled on a consistent number of 50 for all trials, except for the ones where you *know* you're not getting any variation. (Single residue repacking is deterministic.) They had plenty of computational power, so there really wasn't a need to over optimize the number, especially when setting it lower might fail for particular proteins with harder-to-optimize residues. (As alluded to, how convergent the packer is depends a bunch on the context - how the backbone is structured, what the residue identies are that are being repacked, and what are the surrounding sidechains which aren't being packed like.)
Dear rmoretti,
Thank you for your answer! It was very helpful! Yes, I was trying to simulate with the settings of row 3 but based on a protocol given in the DDG manual (https://www.rosettacommons.org/docs/latest/application_documentation/analysis/ddg-monomer).
For now, I have tried both row 3 and row 16. And I worry about a convergence of the results.
For row 16:
1) The first I noticed is that the results (both the repacked structures and estimated ddG values) strongly depend on a crystal structure. In my case I repeat my simulations for two identical chains with slightly different structres in two protien complex's forms, giving me 4 protein structures all together. All of them are quite different, although I might decide to average the numbers.
2) I am wondering how the iteration number of 50 were chosen? My protein consists of 445 residues, which is larger than proteins in the trial set in the Kellogg paper. When I increase a number of iteration I often observe significant changes in the estimated ddG values of the same chain of the same complex form (see a figure attached). Based on the corresponding repacked structures it seems that the results depend on quality of residue conformation sampling, which is not enhanced enough for 50 iterations or sometimes for 200 iterations.
3) I have the same question about avereging of 3 lowest energies for mutant and wildtype. How was the number chosen? Does it show a good convergence of the ddG values?
4) I read many topics about the updated appropriate scorefunction but I am still wondering if these are exactly the settings I have to use (please see below). I could not run talaris2013 scorefunction without setting -corrections:restore_talaris_behavior true.
-corrections:restore_talaris_behavior true
-ddg:minimization_scorefunction talaris2013
#-ddg::minimization_patch <weights patch file>
Thank you for your help again!
Best regards,
Maria.