optH flag makes hydrogen bond energy scores worse…

[Anonymous]

I was trying to run rosetta.score with no_optH = false. However, when compared with the energies without hbond optimization, the hydrogen bond energies get worse, while the other components are unchanged. I have attached the input pdb file below.

Here are the command lines:
rosetta.score -database /usr/local/rosetta/3.3/rosetta_database/ -in:file:s test.pdb -out:output -no_optH true # Turn off hbond optimization
rosetta.score -database /usr/local/rosetta/3.3/rosetta_database/ -in:file:s test.pdb -out:output -no_optH false # Turn on hbond optimization

The output score file is:

SCORE: score fa_atr fa_rep fa_sol fa_intra_rep pro_close fa_pair hbond_sr_bb hbond_lr_bb hbond_bb_sc hbond_sc dslf_ss_dst dslf_cs_ang dslf_ss_dih dslf_ca_dih rama omega fa_dun p_aa_pp ref allatom_rms gdtmm gdtmm1_1 gdtmm2_2 gdtmm3_3 gdtmm4_3 gdtmm7_4 irms maxsub maxsub2.0 rms description
SCORE: -916.848 -1545.330 129.590 764.708 2.831 0.693 -40.420 -43.866 -235.198 -73.020 -51.481 -9.278 0.432 2.847 1.178 -10.334 41.468 264.591 -56.639 -59.620 0.000 1.000 1.000 1.000 1.000 1.000 1.000 0.000 422.000 422.000 0.000 test_0001 # with no_optH = true
SCORE: -911.639 -1545.330 129.438 764.708 2.818 0.693 -40.420 -43.866 -235.198 -69.964 -49.162 -9.278 0.432 2.847 1.178 -10.334 41.468 264.591 -56.639 -59.620 0.000 1.000 1.000 1.000 1.000 1.000 1.000 0.000 422.000 422.000 0.000 test_0001 # with no_optH = false

[Anonymous]

This duplicates in developer trunk; I am looking into it further.

[Anonymous]

Any update on this?

At the least, I would expect that the optimization would return the starting structure and energies, if a better solution cannot be found.

[Anonymous]

I filed a bug report but it hasn’t attracted any interest. We have the big yearly meeting coming up next week so i will try shopping it around there to the packing gurus.

[Anonymous]

Great! Thanks a lot for following up. I will wait until after RosettaCon for any updates.

[Anonymous]

Any updates on this?

Thanks.

[Anonymous]

The official ruling was “well, packing is stochastic and occasionally packing fails to find a better conformation”. Packing isn’t guaranteed to lower energy if it can’t recover off-rotamer atom placements by using only its database rotamers.

I think we should tweak optH to at least check if the energy is increasing and refuse to return the newly worse-packed sidechains, but as it stands it’s considered not a bug but just an insufficiency of Monte Carlo & rotameric sidechains.

[Author]

Posts