Dear Rosetta Users,
I am trying to design a loop present in a protein-protein interface. At the end of my design protocol, I inserted a LoopAnalyzerMover to assess the output decoys. To my dismay, I find very high LAM score, for all loop designs. An example LAM output is below:
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
LoopAnalyzerMover: unweighted bonded terms and angles (in degrees)
position phi_angle psi_angle omega_angle peptide_bond_C-N_distance rama_score omega_score dunbrack_score peptide_bond_score chainbreak_score
pos phi_ang psi_ang omega_ang pbnd_dst rama omega_sc dbrack pbnd_sc cbreak
187 -106.2 151.7 178 1.339 -0.122 0.0155 0.2082.82e+04 0.6
188 -71.4 161.7 -178.2 1.329 -0.811 0.246 0 -3.17 0.0414
189 -130.5 60.51 174 1.329 0.825 0.939 1.86 -3.43 0.357
190 -74.16 166.9 169.8 1.329 -0.599 0.495 0.22 -3.43 1.14
191 -60.36 -16.81 178.3 1.329 -0.73 -0.249 0.977 -3.43 3.24
192 -88.43 165.3 -168.7 1.33 -0.234 2.83 0.139 -3.42 1.2
193 -88.68 -5.647 -171.9 1.328 -0.956 0.572 1.17 -3.42 0.631
194 -123.4 82.26 -178.5 1.328 0.902 0.126 3.04 -3.44 0.169
195 -78.86 156 162.2 1.328 -0.791 2.03 0.0889 -3.44 2.92
196 -83.83 151.7 175.5 1.338 -0.67 -0.000213 0 -3.21 1.87
197 -109.7 121.8 -164.4 1.343 0.0111 3.51 0.7972.82e+04 5.94
total_rama -3.17431
total_omega 10.5213
total_peptide_bond 56327.8
total_chainbreak 18.1063
total rama+omega+peptide bond+chainbreak 56353.2
LAM_total 56353.2
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
As per my analysis, the peptide bond score for the starting and ending residues of the loop seem to be the problem. The loop geometry seems fine to me though. Since, this issue is observed in all the output decoys, I am not sure what is the source of the problem.
Some additional info:
Rosetta Scripts LAM tag:
<LoopAnalyzerMover name="loopAnalysis">
<Loop start="188" stop="196"/>
</LoopAnalyzerMover>
Note that the loop region is from residues188-196 and it is flanked by cysteine residues (187 and 197) that form a disulphide bond. Could this high score be due to the disulphide bond?
Some insights will be very helpful. Thanks in advance.
You're definitely right that it's upset about the peptide bond score.
I like your hypothesis that the disulfide is tripping it up. I looked at the code expecting to find some misused indexing that doesn't account for disulfides/termini/etc, but it appears to just be looking for the N and C atoms directly (I am looking at the file PeptideBondEnergy.cc). It still smells like the disulfides are the problem.
In LoopAnalyzerMover the peptide bond energies are not calculated against the mock poses created to score the chainbreaks - that was my second hypothesis.
The pbnd score is on a gaussian with these settings:
Real const mean( 1.325883 );
Real const sdev( 0.012547 );
The printed distances are like 2 SDs out, clearly not 10K energy units out.
_____
So what should you do? Well, LoopAnalyzerMover exists to highlight things I found wrong when doing loop designs many years ago. It's just meant to raise alarms on things the normal scorefunction doesn't notice (like bond length geometry errors). If LAM isn't working here - ignore it! Reanalyze on only the 188-196 region perhaps. Check out the cart_bonded score as an alternative that will check on bond lengths. I'll file a bug report about this in the meantime, but exactly where the bug is, is not obvious.
Thank you smlewis for the quick response. It seems logical to try and rerun this on the loop excluding the cysteines; will do as you suggested.