You are here

rosetta reproducibility

3 posts / 0 new
Last post
rosetta reproducibility
#1

Hi,
I've been finally able to build rosetta on various machines/compilers. Now I am trying to understand reproducibility/reliability of my builds. I used suggestions from an old post, so I run rosetta -s test.pdb -fixbb -design -natrot on test.pdb from:
rosetta_documentation/docking/tutorial/examples/samplerun/test.pdb

As far as I understood the various options I used, I should got exactly the same results, but I did not. Below is a summary:

compiler gcc gcc icc gcc icc
HW xeon32 xeon64 xeon64 amd64 amd64
score: -159.56 -159.61 -158.66 -158.64 -158.71
env: -37.85 -37.85 -37.85 -37.85 -37.85
pair: -36.99 -36.99 -36.99 -36.99 -36.99
vdw: 0.00 0.00 0.00 0.00 0.00
hs: 0.87 0.87 0.87 0.87 0.87
ss: -157.16 -157.16 -157.16 -157.16 -157.16
sheet: 0.69 0.69 0.69 0.69 0.69
r-sigma: -65.66 -65.66 -65.66 -65.66 -65.66
cb: 47.89 47.89 47.89 47.89 47.89
rg: 15.49 15.49 15.49 15.49 15.49
co: 48.99 48.99 48.99 48.99 48.99
rama: 16.56 16.56 16.56 16.56 16.56
hb_srbb: -12.63 -12.63 -12.63 -12.63 -12.63
hb_lrbb: -87.25 -87.25 -87.25 -87.25 -87.25
bk_tot: -186.77 -186.82 -185.87 -185.85 -185.92
fa_atr: -428.11 -428.11 -428.11 -428.11 -428.11
fa_rep: 41.43 41.43 41.43 41.43 41.43
fa_sol: 194.89 194.89 194.89 194.89 194.89
fa_h2o_sol: 0.00 0.00 0.00 0.00 0.00
fa_dun: 179.11 179.11 179.11 179.11 179.11
fa_intrares: 0.11 0.11 0.11 0.11 0.11
fa_pair: -8.12 -8.12 -8.12 -8.12 -8.12
gb_elec: 0.00 0.00 0.00 0.00 0.00
fa_plane: 0.00 0.00 0.00 0.00 0.00
fa_prob: -35.17 -35.17 -35.17 -35.17 -35.17
fa_h2o: 0.00 0.00 0.00 0.00 0.00
fa_h2o_hb: 0.00 0.00 0.00 0.00 0.00
fa_ref: 14.72 14.72 14.72 14.72 14.72
gsolt: 84.69 84.74 84.62 84.58 84.64
sasa: 6824.47 6826.76 6818.58 6821.43 6821.35
hb_sc: -16.31 -16.36 -15.41 -15.39 -15.46
sasapack: 2.89 2.89 2.90 2.92 2.91

Most of the terms are indentical, whereas some other such as bk_tot, sasa and the total score, are not. Is this discrepancy ok ? I came from the quantum mechanics world, and there numbers between differently built binaries are identical to the 6th digit. Further, I also noticed that some H atoms added during the run to test.pdb have different coordinates in the resulting test_0001.pdb, even by 1 Angs, by using different binaries. Is this ok ? Any explanation ?

Of course, if my binaries have problems, can somebody post the correct numbers ?
Thanks,
Luigi

Thu, 2009-03-05 02:55
lcavallo

Hydrogen placement will depend somewhat on the random number seed - did you try constant_seed in your command line?

Thu, 2009-03-05 13:53
smlewis

Great,
now results match perfectly.
Thanks!
Luigi

Thu, 2009-03-05 14:12
lcavallo