Tue Jul 13 09:00:39 PDT 2004 T0238 DUE 17 Aug Tue Jul 13 11:57:29 PDT 2004 Kevin Karplus The T2K alignment picks up a conserved CYS residue that the t04 alignment misses. This looks like a new fold or difficult fold-recognition problem. Tue Jul 13 12:32:02 PDT 2004 Kevin Karplus Maybe a.118.13 ? It looks like we may be choosing betwen superhelix and bundle structures for the helices. Burial prediction will probably be very important for this. Tue Jul 13 15:41:46 PDT 2004 Kevin Karplus Taking a quick look at the T0238.t2k.undertaker-align.pdb file, it seems to me that we have an all-helical protein that is going to need to be packed, and that the bundles are currently looking more promising than superhelices. Tue Jul 13 16:45:49 PDT 2004 Kevin Karplus The try1-opt1 model is pretty, but far too loose to be correct. I like the section from E100-V185, but I don't know that it is right. There is a possiblity that C22 and C25 are a disulphide, but just as likely is that they form a metal-binding pcket with some HIS or charged residues. We may need to turn off dry12 and phobic_fit on this target, to keep from crumpling the long helices. Sat Aug 7 18:21:25 PDT 2004 Kevin Karplus I haven't looked at this in a long time, but the deadline is coming up, so we'd better do something. I think I'll add the conjectured SSBOND, but weakly. I'll also remove the beta-sheet-forming cost functions, since this seems to be a pure helical protein. Sun Aug 8 12:49:07 PDT 2004 Kevin Karplus try2 looks OK---in fact, it looks pretty good after E100. None of the robetta models look as good to me, but try1 is a bit better. (The try2 costfcn scores try1 better.) The big problem with try2 is that the N-terminal helix is not buried as it is predicted to be. There is a big hole where it might fit. Maybe add some constraints to try to place it there: L6.CG N228.CB I11.CB F236.CG I17.CB I163.CB Wed Aug 11 16:40:25 PDT 2004 Kevin Karplus Jenny selected this target on the 9th, but apparently has not started work on it yet. With unconstrained.costfcn, try3-opt2 scores best, then try2-opt2, try1-opt2. I don't like how the helix breaks neark K81 in try3-opt2, but the helix break at S202 in try2-opt1 also looks bad. I wonder what a run from the existing models would do---perhaps crossover could pick up the good points from both try2-opt2 and try3-opt2. Fri Aug 13 16:29:20 PDT 2004 Kevin Karplus try4-opt2 scores best with an unconstrained costfcn, but is almost identical to try3-opt2. phobic_fit is actually not too horrible as a cost function component, as we are bundling the protein into a fairly compact shape (which only a couple of the robetta models do). I think I'll make a chimera by superimposing try4-opt2 and try2-opt2, taking try2 up to K99 and try4 afterwards. The script for the superposition is try2+try4-superpose.under, the superposed models are try2+try4.pdb, and the chimera is decoys/try2-try4-chimera.pdb This chimera does not score quite as well as try2 or try4, probably because of dry12 and phobic_fit costs, but mainly because it does not have the SSBond I was insisting on for try4 (an easily fixed problem). I'll do an optimization just from the chimera. There is some doubt about whether the N-terminal helix really should be helical---perhaps I should try to make strands out of it?? I'll start a try5 run with strand constraints at the N-terminus, starting from just the try2-try4-chimera. Sat Aug 14 03:10:24 PDT 2004 Martina Koeva Here is something that I thought might be interesting: the output from the SignalP prediction for this target: >T0238 length = 70 # Measure Position Value Cutoff signal peptide? max. C 38 0.521 0.52 YES max. Y 38 0.608 0.32 YES max. S 15 0.992 0.97 YES mean S 1-37 0.822 0.51 YES D 1-37 0.715 0.45 YES # Most likely cleavage site between pos. 37 and 38: ANA-NT >T0238 Prediction: Signal peptide Signal peptide probability: 1.000 Max cleavage site probability: 0.970 between pos. 37 and 38 Sat Aug 14 11:59:05 PDT 2004 Kevin Karplus Hmm, a signal peptide with cleavage between 37 and 38 would not be compatible with a disulfide at C22-C25. The disulfide may be wrong. It is supposed to be an outer-surface protein, which would have an anchor peptide, rather than a signal peptide. (source: http://cassandra.bio.uniroma1.it/Casp6/Targetinfo.html) The try5-opt2 run has some holes, but has the best unconstrained cost so far. We should probably take out the N-terminal strand constraints, since they are not forming strands. Sat Aug 14 15:57:28 PDT 2004 Martina Koeva ID: VS60768 Password: T0238try5 Meanwhile, I took out the strand constraints for the N-terminal residues and am starting a run from existing models that now can include the try5, as well as all others. Or should I try just from try5? Once that is done I will go back and check VAST for any new alignments. Sun Aug 15 05:30:24 PDT 2004 Martina Koeva Try6 is done and it score slightly better than try5 with the unconstrained function. There is still a big hole through the middle of the protein, looking down the helical axis of the larger helices (does that statement make any sense?). I wonder whether going back to alignments and introducing some general constraints to close these large gaps would do any good. On a different note, I put in the comment related to the signal peptide with the idea that a more often than not (according to Voet and Voet) signal peptides seem to take helical form. So I think I like the N-terminus better as a helix than a strand. However, I guess undertaker does not seem to want to fold the N-terminual segment as strands either. I will try 2 things at this point: 1. start from alignments with the new general constraints (try8 on hoot) 2. start from existing models and introduce the new general constraints (try7 on heehaw). Sun Aug 15 14:12:00 PDT 2004 Martina Koeva So try7-opt1 is showing a bit of what I was hoping for with the introduction of the general constraints - the helix 67-96 has bent nicely around to partially close the gap. What I will look for in try9 is to pack helix 192-218, a bit better against helix 160-185. What about the N-terminal helix? It is quite hydrophobic, but does that mean that we need to pack it more toward the core of the protein, like maybe against residues 169-172 on the large central helix? Or maybe we need to leave it out, if it has the function to interact with some non-polar environment outside of the protein? Once try7 finishes, I will start a try9 from existing models and the additional constraints. Mon Aug 16 15:22:11 PDT 2004 Martina Koeva Here is a list of proposed submission candidates: try9-opt2 (best score with unconstrained cost function) try2-opt2.repack-nonPC (lowest Rosetta energy) try1-opt2 (full automated one) T0238-2bidA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 try8-opt2? So of all models that we have, if we exlcude try9 and try1 (since we are submitting models for them already), try7 looks quite like try9 but a tiny bit worse. The general structure and packing though are more or less the same. Try5 also looks very much like try9, but try9 packs the helices a bit tighter. The models that look quite different from the others are try2 and try8. For try2 that seems to be the case, because one the long predicted helices is crumpled into two separate pieces. Tyr8 seems to have overall a different packing, it does not even form a standard bundle (from what I can see). Try8-opt2 also scores better than try2-opt2 with the uncostrained cost function. So even though try8-opt2 is quite far down the score list, I would say for the purposes of diversity, we should go with try8-opt2 as the 5th model. Mon Aug 16 16:23:58 PDT 2004 Kevin Karplus I buy Martina's reasoning for try8-opt2. I'll submit try9-opt2 (best score with unconstrained cost function) try2-opt2.repack-nonPC (lowest Rosetta energy) try8-opt2 for diversity try1-opt2 (full automated one) T0238-2bidA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 Thu Nov 18 23:36:12 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 22.8915 (this was also our model1) best submit 22.8915 (also model1) model1 22.8915 auto 14.1222 align 8.9518 robetta best 19.7286 (robetta model4) robetta1 14.4964