Wed Jul 14 13:57:20 PDT 2004 T0241 DUE 21 Aug Thu Jul 15 10:43:45 PDT 2004 Kevin Karplus Although there is a lot of beta sheet in try1-opt2, I'm not fond of the structure. Some strands are wound into helices (E146-K150) some helices extended into strands (P162-A167) I'd like conserved H174,H176, and H187 to cluster, perhaps as antiparallel strands with H187 ^v I178. probable strands: s1 V4-F6 weak s2 D22-T24 weak s3 M34-T38 very weak s4 E45-M49 s5 K52-I56 s6 K61-E64 s7 F73-H78 s8 K84-T86 weak s9 V109-T126 s9a V109-L115 s9b M121-T126 s10 E146-F149 very weak s11 G154-F159 s12 H174-I178 weak s13 H187-F193 s14 A196-S201 s15 E205-Q209 It looks like S4-S6 forms a reasonable sheet in try1, though the addition of S7 looks suspect SheetConstraint M46 M49 Q55 K52 hbond I47 SheetConstraint Y54 D57 E64 K61 hbond Q55 Let's also add an antiparallel sheet constraint for s12-s13 to cluster the conserved HIS: SheetConstraint H174 I178 L191 H187 Thu Jul 15 17:24:03 PDT 2004 Kevin Karplus try2 scores much better than try1, but there are still just scattered fragments of beta sheets. Y164-N169 should probably be helix, not strand. Thu Aug 12 17:14:51 PDT 2004 Martina Koeva I ran VAST on try2-opt2 and VAST found only one neighbor. VAST ID: VS60694 Password: T0241try2 The hit was relatively weak and it was to 1skoB. I quite like the structural alignment. It points to this configuration of the following strands - s9a, s9b, s11, a new strand s11a, and s12. The alignment implies the following arrangment of those strands: s9b ^v s9a v^ s12 ^v s11a v^ s11 (all antiparallel strands). We already have: SheetConstraint Y111 L115 K123 F119 hbond V113 SheetConstraint G154 F159 N169 Y164 hbond L156 which are combinations: s9b ^v s9a and s11 ^v s11a Oh, but try1-opt2 has all of sheet constraints that I am looking at: SheetConstraint Y111 T116 T179 H174 hbond A112 SheetConstraint Y111 L115 K123 F119 hbond V113 SheetConstraint G154 Y160 N169 N163 hbond L156 SheetConstraint N163 V170 E180 F173 hbond Y164 I will put in those explicitly into the try3.costfcn. I have rescaled the hbond weights, turned down 'sidechain'. Sat Aug 14 02:32:11 PDT 2004 Martina Koeva I have started try4 with the same sheet constraints, but introducing rr constraints. I will comment more in the morning, but I envision this structure forming a full sheet of strands with helices packing on both sides of the sheet. Sat Aug 14 15:39:17 PDT 2004 Martina Koeva Try4 wraps the second sheet around against the first one. There is still one strand that is folded into a helix and I am wondering whether it should not replace the strand that is predicted to be a helix, but is in the sheet right now. However, try4 is a better structure and i am submitting it back to VAST to see whether I might not be able to find some clue as to how the structure should fold. VAST ID: VS60767 Password: T0241try4 Sun Aug 15 16:14:56 PDT 2004 Martina Koeva One of the paths that I want to pursue is to try and build a one whole sheet from all strands. In try4 (as well as earlier ones)we look like we have two separate sheets. For try5,I'd like to introduce the following sheet constraint: SheetConstraint F73 H78 A124 F119 hbond F73 7 The purpose of this constraint is to join in an antiparallel fashion the two "edge" strands on the current sheets. Based on str2 t04 predictions, both of those strands are predicted to have an antiparallel partner on each side, so such an attempt should not be completely unreasonable. Mon Aug 16 18:05:32 PDT 2004 Martina Koeva I am starting a try6 in an attempt to replace predicted helix 163-170 with the strand 186-194 in the large sheet that we are forming in try4. For these purposes I have put together the following sheet constraints: SheetConstraint G154 Y160 Q192 G186 hbond L156 7 SheetConstraint G186 F193 E180 F173 hbond H187 7 Wed Aug 18 00:32:24 PDT 2004 Martina Koeva I just updated the Makefile with the MANUAL_TOP_HITS and ran make extra_alignments, as well as updated the all-align files. I am going to start another run from alignments as try7. For now I will just leave the constraints as they are and just increase the weights. I have also updated the dry/wet/near_backbone/way_back weights to higher values (as noted during our Monday meeting 08/16/04). I have also increased phobic_fit from 2 to 4. I just noticed a bad typo in the .costfcn files after try3. I must have mistyped something and instead of changing dry6.5 from 6 to something different, I had changed it into dry6.15 and had left the weight of 6 the same. I am correcting that for try7.costfcn. Argh... First, though, I also noticed that this directory never got its t04.many.frag files, so I am redoing the initial make so that the file is created. I will uncomment the PrintRemplateAtoms Template.atoms line, which I should not forget to comment back out in try8. After try7 is done, I want to do a run from existing models, which will be try8. Wed Aug 18 01:14:12 PDT 2004 Martina Koeva Yay! t04.many.frag file has been generated and I am starting try7 on ribbit. Wed Aug 18 13:44:10 PDT 2004 Martina Koeva Try7 completely ignored my sheet constraints, so for try8, instead of starting from existing models, I will start from alignments again, take out the hbond from the sheet constraints, since I really do not know how the strands are supposed to line up. I have increased the sheet constraint weights, as well as some of the strand constraint weights. My goal for try8 is still to bring the strand 186-193 down into the sheet to replace the predicted helix that is currently (in try7-opt2) back in the sheet. Wed Aug 18 17:13:52 PDT 2004 Martina Koeva Try8-opt1 looks like it's trying to pull down the predicted strand into the sheet as I requested. It also shows a larger sheet forming in the N-terminal part of the protein (we are forming strands that we don't predict to be strands, but they are also forming sheets). I will have to wait though for try8 to finish, after which I will run try9 from existing models. Interestingly enough, the uncostrained function seems to like the try4 and try7 much better than try6 and try8-opt1 models. Try4 and try7 have the predicted helix inserted in the sheet, fitting more with the alignments for this target: T0241.try4-opt2.pdb.gz 171.9556 T0241.try7-opt2.pdb.gz 173.7664 T0241.try6-opt2.pdb.gz 184.9558 Try8-opt1 is way down - it scores the worst from all of the try*-opt1 models. --------------------------------------------------------------------------- From martina@soe.ucsc.edu Wed Aug 18 17:11:51 2004 MIME-Version: 1.0 Date: Wed, 18 Aug 2004 17:11:50 -0700 (PDT) From: Martina Koeva To: Kevin Karplus Subject: T0241 In-Reply-To: <200408182344.i7INihxQ016622@cheep.cse.ucsc.edu> Hi Kevin, I have a quick question about T0241. I have been struggling a little bit with it in figuring out what I want the fold to look like. We are forming a lot of sheets, but I am having a hard time figuring out how they should go together. I looked at the domain predictors and they seem to predict domain breaks at various places, including residues 22,72,76,127 and 134, none of which seem plausible to me. I was thinking that a possible domain break could come in around residue Val90. Do you think that it would be worth, splitting the protein into 2 domains and folding them separately? If I set up those tonight, there might be enough time to put them back together (T0241 is due this Saturday). What would your suggestion be? The latest structure that we have for T0241 is try8-opt1 (try8 is still running). Thank you! Martina ---------------------------------------------------------------------------- From karplus@soe.ucsc.edu Wed Aug 18 17:57:07 2004 Date: Wed, 18 Aug 2004 17:57:06 -0700 From: Kevin Karplus To: martina@soe.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: (message from Martina Koeva on Wed, 18 Aug 2004 17:11:50 -0700 (PDT)) Subject: Re: T0241 Both the N-terminal region and the C-terminal region seem to be trying to form anti-parallel sheets. Whether these are part of the same sheet or two separate sheets is hard to tell. The fold-recgonition hits are all very dissimilar---mainly picking out pairs of strands in different folds. I think you should pay attention to H174, H175, and H187. They are strongly conserved and almost certainly cluster. You could add some disance constraints assuming that they coordinate a metal ion. The strand and helix predictions for t04 seem pretty strong, so you might want to weight the T0241.to4.dssp-ehl2.constraints highly---at least 10 times what you are giving the rr constraints. The long predicted strand from V109 to T126 may well be 2 strands, or it may be the connector that holds two sheets together. I don't have any great ideas on T0241---it may be that folding up a few unrelated sheet pieces is as good as we can currently do. Incidentally, don't worry too much about the unconstrained costs---to a large extent they will just represent breaks and clashes when we are this far from a real solution. ----------------------------------------------------------------------- Thu Aug 19 13:14:59 PDT 2004 Martina Koeva At Kevin's suggestion, I have introduced distance constraints (starting try9) so that we can have the conserved histidines clustered.I have also increased for try9 the weight on the helix and strand constraint ten-fold. Try8-opt2 shows the formation of a larger N-terminal sheet, which almost twists like a barrel, except that there are strands, going outside of it too. Unfortunately, the strand that I was looking for to come down into the sheet has not done that completely. There are some bad breaks and after try9 and try10 are done, I will have to start increasing the break weights. Try9 includes the above noted changes and starts from alignments again, while try10 has the same cost function as try9, but starts from existing models. Thu Aug 19 15:57:20 PDT 2004 Martina Koeva Hmm, I just looked at try9-opt1 and things don't look very good. If we try to cluster the histidines, undertaker seems to like the idea of the predicted strand that has H187 not being in the C-terminal sheet. In that case, it tries to put the predicted helix 163-169 back in the sheet. The problem occurs though that H174 and H176 are pointing away from H187. The histidines on the different strands seem to be pointing in opposite directions, so something is obviously wrong, if they actually really do coordinate a metal ion. I wonder what would happen, if the strand containing H187 was in the sheet, pointing in the same direction as H174 and H176, but I cannot seem to be able to get undertaker to align them. I will have to wait and see what try10 does with the cluster of histidines. The two histidines - H174 and H176 look like they want to cluster/coordinate with E139, but E139 is not well-conserved. I am not quite sure what to look for. I just went through all the models that we have (opt2 models) and try8-opt2 is the only one that we have that seems to have the histidines pointing at least in the same direction, even though they seem to be quite far away from each other to be able to coordinate a metal ion. Thu Aug 19 16:20:40 PDT 2004 Martina Koeva Instead of waiting for try9 and try10 to finish, I will start a try11 from alignments, where I will take out all sheet constraints and leave just the distance constraints for the clustering of the histidines. I looked at the Robetta models and some of them (particularly model 1) look like they are clustering the histidines quite well. It does look like a metal-coordinating cluster. It would be nice, if we can get at least one model that has the cluster, even if it does not do well on forming sheets. Thu Aug 19 16:27:12 PDT 2004 Martina Koeva For try11 I have taken out all sheet constraints related to the C-terminal sheet. I have left in the N-terminal sheet constraints, since they should not have to affect the histidine cluster. Fri Aug 20 03:02:58 PDT 2004 Martina Koeva Try9,10 and 11 are done. I have started try12 (from try10) and try13 (try11) retaining the majority of the cost functions, but raising in both the break and soft clashes weight. I have also raised the dry 6.5 weight from 20 to 25. Both try12 and try13 are starting from existing models. Try11 managed to cluster the histidines to some extent, but they are still not pointing in a direction so that they look like metal-cooridnating residues. I am not quite sure what to do. Maybe the distances that I am using are wrong. I have increased the weight on those distance constraints and am hoping that the breaks introduced in both try10 an try11 on the strand holding H187 will close and help with the repositioning of the histidine. After the runs are done in the morning, I will have to examine the models that we have so far and look into selecting models for submission. Fri Aug 20 11:41:13 PDT 2004 Martina Koeva Try13 looks horrible. The strand with the two histidines has curled up into a helix and all histidines point in different directions. Try12 looks semi-decent in its sheets, except that the histidines do not cluster, despite the weight. I am not quite sure what to do at this point: I might increase the break weight even more, since try12 does not close the break in the strand above the C-terminal sheet. If we were to submit now, I would say: try12-opt2 best unconstrained try13-opt2.repack-nonPC lowest Rosetta energy try1-opt2 full auto T0241-1s1dA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 try11-opt2(?) Maybe as a 5th model we should have a model that at least clusters the histidines. The closest that we get seems to be try11-opt2, but otherwise it looks terrible - breaks, secondary structure. Another option for a 5th models is one that attempts to put the predicted strand near the C-terminal sheet into it and tries to take out the predicted helix out of the sheet. We never really succeed at doing that. The closest that we seem to get is maybe try8-opt2, which has quite a few breaks too. I am going to start a try14 from try12, increasing the break weight even further. Fri Aug 20 12:14:36 PDT 2004 Kevin Karplus Martina just overwrote the notes I had made about try12 in the README file. I'll have to try to remember them. try12-opt2 scores best with the unconstrained costfcn, and the two sheets look pretty good, but the C-terminal part, Q192-K237, looks misplaced. I suspect that part of it is an edge strand for the second sheet, and the helix probably packs against the sheet. The conserved histidines do NOT cluster in try12. H174 and H176 are naturally close, but on the opposite side of the sheet from H187. One problem is that Martina has been clustering the CA atoms of the histidines, but it is the ND1 or Ne2 atoms that we want close. I'll make another cost function that asks the histidines to cluster: his-cluster.costfcn. Fri Aug 20 12:41:06 PDT 2004 Martina Koeva Argh, I uploaded the version of the README after I updated it and it was probably during that time that Kevin changed it. I am sorry about that. I have now attempted to use vi to change README files (when logged in remotely). Fri Aug 20 12:47:56 PDT 2004 Kevin Karplus I created a new try15 costfcn that has the helix and strand constraints plus the his cluster. It likes some of the ROBETTA models best, so we aren't even competitive with robetta yet. I notice that Martina still has not set up her paths appropriately for the compbio tools. In try14.log, all the scwrl runs are aborted because she doesn't have scwrl in her path. [Martina responds, I had my paths set to include the compbio directories and the last time I checked they were working fine. However, I have had one change made in the last two weeks and now it seems none of the export commands in my bash file seem to be executed. ] Fri Aug 20 13:09:11 PDT 2004 Kevin Karplus One reason that try15 may be favoring robetta models so much is the large break weight. For try16, I'll try reducing the weights of breaks and soft_clashes, so that slightly broken models can compete. Now try11-opt2 slightly beats robetta-model2. I'll start a run for try16 also. Fri Aug 20 14:03:46 PDT 2004 Kevin Karplus try15-opt1 is done, and it seems to be mainly robetta-model2 with the histidine rotamers changed and the subdomains packed differently. SCWRLing it throws away 14.6 points on the constraints, though that still beats try11-opt2.repack-nonPC, the next best on the try16 costfcn. Try15 did look at several different models on the first round, and settled on the robetta-model2 one. It looks like try16 will polish try11-opt2.repack-nonPC to a similar level. From martina@soe.ucsc.edu Fri Aug 20 14:12:31 2004 Date: Fri, 20 Aug 2004 14:12:29 -0700 (PDT) From: Martina Koeva To: karplus@soe.ucsc.edu Subject: T0241 try15 favoring Robetta models In-Reply-To: A few of the Robetta models cluster the histidines really well (like Robetta model 1, if I remember correctly), even if they do not form nice sheets. On the other hand, try11 is the only one that we have that seems somewhat close to clustering them (I had completely overlooked the fact that we needed the ND1 and NE2 atoms - no wonder I was not getting results). Could that be another reason why try15 is favoring the Robetta models? From karplus@soe.ucsc.edu Fri Aug 20 14:45:05 2004 Date: Fri, 20 Aug 2004 14:45:03 -0700 From: Kevin Karplus To: martina@soe.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: (message from Martina Koeva on Fri, 20 Aug 2004 14:12:29 -0700 (PDT)) Subject: Re: T0241 try15 favoring Robetta models It is the clustering of the histidines that try15.costfcn and try16.costfcn were set up to detect, so it is not surprising that they favor robetta-model2 and try11-opt2. ============================================================ Fri Aug 20 16:48:08 PDT 2004 Kevin Karplus Everything here looks like trash, but I'll submit anyway try15-opt2 best HIS cluster (mainly from robetta-model2) try16-opt1 another HIS cluster (polished from try11-opt2) try14-opt1 best unconstrained try8-opt2 HIS on same side of sheet but not clustering try1-opt2 full auto I'll replace models 2 and 3 by try16-opt2 and try14-opt2 in the morning. Fri Aug 20 17:44:06 PDT 2004 Kevin Karplus try16-opt2 send as model2. Thu Nov 18 23:41:13 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 11.1278 (try10-opt2) best submit 10.2847 (also model1) model1 10.2847 auto 9.0830 align 5.0953 robetta best 11.0357 robetta1 7.5968