Wed May 31 09:37:25 PDT 2006 T0306 Make started Wed May 31 09:38:03 PDT 2006 Running on lopez.cse.ucsc.edu Wed May 31 11:45:46 PDT 2006 Kevin Karplus No good blast hits (best is E-value 0.3 for 1pixA) Multiple alignments contain no PDB sequences. Scores with HMMs also looking poor, both for target models and template models. Wed May 31 14:41:33 PDT 2006 Kevin Karplus Best E-value is 1.3 for 2ae0X. There is no apparent consistency in the top hits, so this really is ab initio. Wed May 31 16:38:50 PDT 2006 Kevin Karplus The undertaker alignments are not very consistent, though they each pick up some beta sheet. The try1-opt2 has some nice beta sheet, but I don't like the N-terminal helix (which was predicted to be strands). The try1 run seems to have picked up stuff from 2ae0X alignments and then from 2gw6A alignments. We may need to create SheetConstraints by hand for the first 2 strands, if we can't find an alignment that has them. Mon Jun 19 17:02:29 PDT 2006 Kevin Karplus This one looks very difficult, as all the strands appear to be buried, but there is nothing to bury them with. The protein is probably not well structured as a monomer, but only as part of some larger complex. It appears to be labeled as a structural protein of the carboxysome. PMID: 16081736 Science. 2005 Aug 5;309(5736):936-8. Protein structures forming the shell of primitive bacterial organelles. Kerfeld CA, Sawaya MR, Tanaka S, Nguyen CV, Phillips M, Beeby M, Yeates TO. Bacterial microcompartments are primitive organelles composed entirely of protein subunits. Genomic sequence databases reveal the widespread occurrence of microcompartments across diverse microbes. The prototypical bacterial microcompartment is the carboxysome, a protein shell for sequestering carbon fixation reactions. We report three-dimensional crystal structures of multiple carboxysome shell proteins, revealing a hexameric unit as the basic microcompartment building block and showing how these hexamers assemble to form flat facets of the polyhedral shell. The structures suggest how molecular transport across the shell may be controlled and how structural variations might govern the assembly and architecture of these subcellular compartments. It looks like we have 2a10A (a hexamer of wedge-shaped units) that may give us an idea what sort of thing we are looking for. It did *not* score at all well with any of our HMMs and the secondary structure doesn't match our predictions, but it does show how a beta sheet can be buried on both sides in a multimer, but not a monomer. We also have some fairly strong indication that we want to make a hexamer. Mon Jun 19 18:59 PDT 2006 Zack Sanborn Chris and I constructed a model using Proteinshop that tried to follow the topology laid out during our Targets Meeting today. We initially started by flattening all sheets (decoys/Proteinshop/model1.pdb), but then we decided to try to make a wedge with it, so that it would form a circular multimer as described in the paper cited above. The results of the ProteinShop-ing can be found in decoys/Proteinshop/model2.pdb). This model was renumbered using Firas' script to give us the PDB file decoys/Proteinshop/model2_renumber.pdb. An unconstrained optimization of this structure is being run on camano started at 19:04 (try2). It appears to be working. I wasn't able to get sheet constraints from the Proteinshop model, it would just make a blank file using PrintConformSheets. I'm hoping the optimization will make a better structure and give me some constraints. Mon Jun 19 21:00 PDT 2006 Zack Sanborn The optimization run (try2) has finished. Unfortunately (but expected) undertaker didn't like the ProteinShop model we came up with. However, the new structure (try2-opt2) is the highest scoring model (by a little bit) and the structure now fits the secondary structure prediction much better. But, looking at the structure in space-fill reveals some gaping holes in the middle of the structure. So, not likely a good model, but maybe with some refinement, it will pack a little better. Whoops, I may have found the problem. I did not comment-out some lines, so it read in some fragments from *.undertaker-align.under. Not sure if this would have caused the problem or what. I've commented out and am rerunning the optimization (hopefully) just using the ProteinShop model. try3 started on lopez at 21:12, optimizing just the ProteinShop model (model2_renumber.pdb). Mon Jun 19 22:20 PDT 2006 Zack Sanborn Optimization run try3 has finished and BOY is it odd. Undertaker has taken all of the beta sheets and turned them into helices! This must have something to do with the fact that no sheet constraints were used in the cost function. So, undertaker did what it wanted. The model it produced (try3-opt1 and try3-opt2) did not score well relative to the try2-opt* models. So, try2-opt2 is the model that should be refined given that it is the closest to the secondary structure prediction out of all of the models thus far. I will start another runs starting from all models as a polishing run and see where that gets us. I think we should do another run starting from the ProteinShop model with the sheet constraints in place. I just don't know to make them quite yet. Chris? Edit: The run above is try4. JZS. Tue Jun 20 14:49:50 PDT 2006 Chris Wong I've finally been added to the "protein" unix group! Grant and George have suggested that I try to run undertaker again using 2a10A and 2a18A as "top hits". I started out by adding the alignments to these chains. This is done by adding MANUAL_TOP_HITS to Makefile and then doing: 1) make extra_alignments 2) make read_aligments Next, I made a try5.under file based on try1.under. I added lines 80 to 84 to tell undertaker to use those additional alignments. I also copied try1.costfcn as try5.costfcn and removed the constraints at the end of the file. Finally, I start undertaker going with: (make -k T0306.do5 >& do5.log ; gzip -9f do5.log) & The job is running on camano. (15:00) Tue Jun 20 17:03:36 PDT 2006 Chris Wong While try5 was running, I took a look at try4. It looks like try4-opt2 has the secondary structure that we were trying to build yesterday in ProteinShop, except for residues 2-5 which are predicted to be sheets but is modelled as helix. The other thing is that there is still a lot of space inside the model. In some areas, it's possible to see through to the other side of the model. Is it possible to use contact constraints to close these holes up? The results of try5 look too much like 2a10A and 2a18A. Some of our sheets have turned into helices. I don't think the try5 models will be very useful, but at least I learned how to use 'MANUAL_TOP_HITS'. Tue Jun 20 22:35:29 PDT 2006 Kevin Karplus If you really want a model based on 2a10A and 2a18A, you should move the read-alignments-scwrl lines before the first TryAllAlign, and comment out anything else that provides long alignments: Include T0306.t04.undertaker-align.under Include T0306.t06.undertaker-align.under Include T0306.t2k.undertaker-align.under Include T0306.undertaker-align.under ReadFragmentAlignment NOFILTER SCWRL all-align.a2m Since you have not told me which models to submit, and I have to submit tonight (to make the soft deadline), I'll pick some rather arbitrarily. We'll need to do a more careful job before the hard deadline. unconstrained.costfcn: try5, try4, try2, try1, try3 try5.costfcn: try5, try4, try2, try1, try3 try1.costfcn: try4, try2, try5, try1, try3 try3.costfcn: try5, try4, try2, try2, try3 OK, it is clear that no one likes try3, and that try4>try2>try1, and the only difference is where try5 should go. Let's see about putting it third. Tue Jun 20 22:46:51 PDT 2006 Kevin Karplus Unfortunately, try4 and try2 are almost indistiguishable when superimposed, so I'm going to drop try2 and add in one alignment (perhaps to 2ae0X, which was used for try2?) Tue Jun 20 22:57:36 PDT 2006 Kevin Karplus Preliminary submission of ReadConformPDB T0306.try4-opt2.pdb ReadConformPDB T0306.try5-opt2.pdb ReadConformPDB T0306.try1-opt2.pdb ReadConformPDB T0306.try3-opt2.pdb InFilePrefix ReadConformPDB T0306.undertaker-align.pdb model 1 done. Wed Jun 20 12:18 PDT 2006 Zack Sanborn I started a try6 run on camano at 12:17. This run is redoing try5 that is trying to align to 2a10A and 2a18A, but I have commented out the reading in of long alignments and moved the read-alignments-scwrl before the first TryAllAlign as per Kevin's suggestion. Side note to Kevin, sorry about not telling you which were the best models last night for the submission. A little bit of miscommunication between Chris and me. Whoops, screwed up something small, but caught it looking at the logs. Restarted try6 on camano at 12:32. Wed Jun 21 16:18 PDT Zack Sanborn try6 had finished, but didn't score as well as the try5-opt2. It scores second best, similarly to most of the models that score lower than it. But, try5-opt2 has a much lower score than the rest. The structure looks good, but still isn't exactly what we wanted from our initial Proteinshop model. I believe the reason for this is that we don't have any sheet constraints that are forcing the structure we want. Now, we've attempted to figure out some manual sheet constraints to force the beta sheets we wanted based on our initial Proteinshop'd model. The sheet constraints are given below: SheetConstraint (T0306)L3 (T0306)V6 (T0306)V14 (T0306)V11 hbond (T0306)A4 1 SheetConstraint (T0306)V14 (T0306)V11 (T0306)A41 (T0306)I44 hbond (T0306)T13 1 SheetConstraint (T0306)V55 (T0306)S59 (T0306)V78 (T0306)V82 hbond (T0306)L56 1 SheetConstraint (T0306)V78 (T0306)V82 (T0306)I92 (T0306)G88 hbond (T0306)I79 1 Inputting these constraints into the try7.costfcn, and making a try7.under that is only using the Proteinshop model, we started try7 on lopez at 16:20. Wed Jun 21 20:44 PDT 2006 Zack Sanborn Well, try7 has finished and try7-opt2 is now the highest scoring model by a big margin, likely because our model gets the sheet constraints mostly right (we did, of course, get the sheet constraints by looking at this model, so this only makes sense). It looks like it kept the overall topology of the Proteinshop model, but to get some of the sheet constraints (i think the second one from the top in above sheet constraints) it had to break the chains. So, I'm doing a try8 that starts with all models (including the Proteinshop one) with increase costs for gaps (doubled) and the "offending" sheet constraint reduced relative to the others. Interestingly, we were able to make the beta sheets out of the top layer. The layer of beta sheets beneath these were left as coils. It might also be helpful to go back into Proteinshop and push the strands closer together (making small clashes) so that Undertaker can more easily optimize it while being more likely to force the structure we want. We should also take another look at the sheet constraints if try8 doesn't work too well. try8 started on camano at 20:42. Wed Jun 21 23:47 PDT Zack Sanborn try8 has finished and it looks (at first glance) to be unchanged from try7, although it does score better. I think we need to change the constraints and/or fix the bottom 4-strand beta sheet (right now a set of broken coils) using ProteinShop. Fri Jun 23 12:32 PDT 2006 Zack Sanborn I've done another ProteinShop model, this time trying to make the strands closer together (getting some small clashes) with the hope the Undertaker will be able to make a better model from it. The new model is found at /decoys/Proteinshop/model3_renumber.pdb. I've started a run (try9) that starts from this model and applies the same sheet constraints. It was started on camano at 12:30. Fri Jun 23 15:51:14 PDT 2006 Chris Wong try9 has the sheets structure that we were trying to achieve with the ProteinShop modelling. However, there is a short coil/loop section that has breaks on either end. We'll do a try10 with unconstrained cost function and increased break penalty to try to polish away the breaks. (make -k T0306.do10 >& do10.log ; gzip -9f do10.log) & started on lopez at 3:54p. Sat Jun 24 21:27 PDT 2006 Zack Sanborn Well, we have our desired structure. But, I worry that the flatness of the sheets isn't very protein-like. So, I'm taking the sheet constraints and helix constraints from this model and starting again from the original alignments (essentially copying the try1.under script, but commenting out the PrintTemplateAtoms line). try11 was started on camano at 21:25. Mon Jun 26 12:05 PDT 2006 Zack Sanborn Well, try11 finished. It definitely looks more protein like than the try10 models, which of course was expected given that we weren't starting from the ProteinShop model. Thankfully, it also kept most of the desired topology. We lost most of the bottom layer of sheets (part of them turned into helices), but the top layer of sheets and the helix were mostly maintained as we wanted. I'm running an unconstrained optimization on this structure to tighten up the breaks and try to lower the soft clashes. I also want to see how this structure will hold up when the sheet constraints are removed. Mon Jun 26 14:07 PDT 2006 Zack Sanborn try12 finished. It now scores about 4th or 5th best. try5, try4, and try6 score slightly better than try12. I'm running try13 to try to tighten up the structure a bit more. try13 started on camano at 14:08. Mon Jun 26 17:23 PDT 2006 Zack Sanborn try13 finished and is now third best, scoring approximately the same as try6-opt2. However, try5-opt2 still scores better. I've decided that I don't like the helix around Ala66 - Glu70. I think it should be longer. Also one of the unsatisfied sheet constraints is causing a funny short beta-loop-short beta structure. I think if we remove the following constraint, Undertaker might find something more sensible to do with this structure. try14 that extends the helix constraint and removes the described sheet constraint was started on orcas at 17:22. Tue Jun 27 19:22:26 PDT 2006 Chris Wong The helix containing Ala66 - Glu70 has been extended, but a break has been introduced between S72 and P73. It seems like it is very difficult to come up with a model that doesn't have a break. try13 had a break between E53 and W54. The sheet between W54 and G60 looks odd to me. It doesn't really line up with any of the other sheets. In fact, the 3 sheets from W54 to the end of the protein look like they're out of alignment. Wed Jun 28 12:36:53 PDT 2006 Zack Sanborn So far, our best scoring model from the ProteinShop-borne model is try14-opt2, which scores third best to the top scoring models try10-opt2 and try8-opt2. However, this model has a sizeable break after the helix between residues Ser72 and Pro73. We are trying two approaches to fix this. The first is to start with the try14-opt2 model and increase the penalty for breaks and gaps to see if undertaker can heal that break somehow. Our worry is that it may rip apart our nice structure to do this. So, in case this happens, we are trying to heal this gap by remaking the try14-opt2 run with slightly different constraints. We have loosened the HelixConstraint in the try14.costfcn to make a smaller helix. It is hoped that the smaller helix with more coil on either side of it will have enough play to not cause the break. try15, the run starting from try14-opt2 and doubling the penalty for gaps and breaks, was started on shaw at 12:55. try16, the run remaking the try14 model with altered constraints, was started on camano at 12:30. Wed Jun 28 16:03:44 PDT 2006 Chris Wong try17 was a polishing run on try16. It is unconstrained. try15 has a break between S72 and P73. It appears that we can't close the break by simply raising the break penalties. try16 has no breaks. We are more successful in closing the break by changing the helix/sheet constraints. The kink in the helix at the beginning is removed, compared to try14 and try15. try17 has no breaks. Actually, L76 and C77 are missing when viewing in rasmol! They are not missing when viewing in pymol. How/Why does this happen ? Wed Jun 28 17:29:06 PDT 2006 Kevin Karplus try17 < try16 < alignment to 2gw6A, which is the same template used in try1-opt2. try5 < alignment to 2a10A Wed Jun 28 17:43:05 PDT 2006 Chris Wong We have three paths to take on this one. 1) models of the family try17. These models can be polished further, removing disulfides and metals from the costfcn. The break at L76 and C77 needs to be repaired, too. Maybe this one can be optimized as a hexamer. 2) models of the family try 5. These models can also be polished without the disulfides and metals in the costfcn. Again, need to polish this one as a hexamer. 3) models originating from extensive ProteinShopping. More ProteinShopping can be done to create a wedge with the proper specifications for forming a hexamer. note: a hexamer can be formed in two ways, s(2,3) and s(6). In either case, we should ProteinShop the model so that the hydrophobic face forms a 60 degree angle. Also, the hydrophilic residues need to be on the outside face of the monomers. Before further ProteinShopping, we are going to run try10-opt2 through VAST to see if there's anything that looks similar in the databases. VAST results: 4 neighbors found. 1) 1SEOb - 57 aligned residues 2) 1k2xb - 50 aligned residues 3) 1srsa - 14 aligned residues 4) 1sz6a - 13 aligned residues Thu Jun 29 15:16:42 PDT 2006 Chris Wong 1) Did a little bit of work with Proteinshop (model4.pdb). What I was trying to do here was to move the helix and sheets over a little bit to make a roughly 60 degree angle on the other side of what we had in try10-opt2. 2) Took another look at 2a10 and 2a18 structures (around 105 residues each, our target is 95 residues). These are the carboxysome hexameric structures. I think our protein might look like either of these. What's similar between them is that they are s6 hexamers with the sheets pointing (more or less) away from the center of the hexamer. I think it might do us good to do an undertaker run forcing 2a10 and 2a18 as top alignments. This was done previously with some mistakes (try5, June 20). Thu Jun 29 18:59:14 PDT 2006 Chris Wong I'm going to try to re-run try5 as try18. I'm doing this because I made some mistakes in try5. (make -k T0306.do18 >& do18.log ; gzip -9f do18.log) & started on camano at 19:21 Fri Jun 30 12:40:30 PDT 2006 Chris Wong I took a look at try18-opt2. Now, it looks similar to one chain of 2a10, except that we have 2 beta strands at the end instead of two helices. From looking at 2a10, it looks like that part of the hexamer is for "grabbing" neighbor chains. Our plan for this guy is to use ProteinShop to build a hexamer out of this thing, and then optimize the hexamer. Fri Jun 30 16:46:12 PDT 2006 Chris Wong We build several hexamers today. 1) decoys/ProteinShop/T0306.try18-opt2_6mer_renumber.pdb is try18-opt2 proteinshopped into a hexamer. try18-opt2 was created by using 2a10A and 2a18A alignments. 2) decoys/ProteinShop/model4_6mer_renumber.pdb is a hexamer that was proteinshopped from our proteinshop model3. 3) Zack made some multimers using a script. (Zack, can you record what you did here?) The next step is to optimize the hexamers. How do we do this? I am going to guess that this is not the same as optimizing monomers. Fri Jun 30 21:53:35 PDT 2006 Zack Sanborn Firstly, the hexamer that I made was done using the make-6mer.under script. This used the hexamer template 2a10 to build a hexamer from the monomeric model. The monomeric model that was "hexamerized" was try18-opt2 that was also built from the 2a10 alignment. Unfortunately, this alignment did not score well, so I don't have a good feeling about this model. Looking at the hexamer generated (6mer/decoys/T0306-6mer-2a10.pdb) you see a large hole. However, most of the hydrophobics are buried. The next step is to optimize these structures. I have *some* experience optimizing multimers (from target T0308), so I hope I do it right. Before starting the optimization, I've copied the hexamer models that we're going to optimize into the 6mer/decoys directory. I will start 3 optimizations, one for each multimer model: try1: T0306.6mer-2a10.pdb (Generated by Script) try2: T0306.model4_6mer_renumber.pdb (ProteinShop'd) try3: T0306.try18-opt2_6mer_renumber.pdb (ProteinShop'd) Each of these use the "multimer 6" option in OptConform to make cyclic hexamer where all monomeric models are forced identical. Also, in the cost function the following known breaks are inputted to tell Undertaker where the monomers begin/end: KnownBreak M96 KnownBreak M191 KnownBreak M286 KnownBreak M381 KnownBreak M476 I've started try1 and try2 on camano, and I started try3 on Sun Jul 2 19:13:10 PDT 2006 Zack Sanborn Hmmm, for some reason I failed to complete the above sentence. try3 was started on lopez. All jobs have completed, however try2 died almost immediately with the following failed assertion: undertaker: XYZpoint.h:57: void XYZpoint::unitize(): Assertion `mag>0' failed. Not sure what caused it, although it did happen on the pure ProteinShop model, so ProteinShop is probably the culprit (gotta love that scapegoat!). try1 and try3 completed normally but I haven't had a chance to look at the models produced yet. Mon Jul 3 17:28:41 PDT 2006 Chris Wong Taking try3-opt2 model, made a new hexamer. This time, we try to rotate slightly to bury some hydrophobics that were mostly on the helices. The new hexamer is called T0306.try3-opt2_6mer_renumber.pdb. We are optimizing this model (reducing the num_gen and super_num_gen so it won't take forever). ( make -k T0306.do4 > & do4.log ; gzip -9f do4.log ) & started at 1730 on orcas. Tue Jul 4 11:37:47 PDT 2006 Chris Wong try4 completed ! It looks a lot like try3-opt2, except that the antiparallel sheets at the end of the protein seem to fit into the multimer structure a little better. Wed Jul 5 14:08:29 PDT 2006 Zack Sanborn I agree with Chris, the try4 structure does look pretty good. The hydrophobics are hidden a little better and its overall structure has a bowl-like structure, similar to the 2a10 template model we're trying to emulate. However, I feel like it could be packed a little better. So, I'm taking the try4-opt2 model and optimizing it, increasing the weights for "phobic_fit" and "dry5" to make it pack a little better. try5 was started on orcas. Fri Jul 7 14:16:17 PDT 2006 Zack Sanborn Ooops, Chris and I forgot to optimize the monomeric models. He's currently busy working on making another hexamer model, so I'm going to take care of these monomers. There are two models we want to optimize, try17-opt2 adn try18-opt2. try17-opt2 came from try16-opt2 that was based on try1-opt2 of all models. try18-opt2, on the other hand, is a re-run of try5, which was making a model based off an alignment to 2a10A (a weakly scoring template but is a carboxysome like this protein). Both optimization runs will be polishing runs where breaks, clashes, dry5, etc. have been increased. try19 is optimizing try17-opt2 and was started on orcas. try20 is optimizing try18-opt2 and was started on ... Fri Jul 7 15:02:02 PDT 2006 Chris Wong We just finished proteinshopping a new hexamer. This is one is T0306.try5-opt2_6mer_renumber.pdb. To make this one we took the following steps: 1) Make a monomer from try5-opt2 2) Proteinshop half of a 6mer. 3) Proteinshop a 6mer out of 2 3mers. What's different about this model is that the helices and sheets alternate, going around the ring of the hexamer. Note: the antiparralel sheets at the end of the protein are quite exposed on this model. We may have to do something about that section at some point. This newest 6mer will be optimized as try6. started on vashon at 1500. Sat Jul 8 08:00:20 PDT 2006 Chris Wong try6 is done. It scores 9th. This is comparing the hexamer scorings. It looks like try6 is gaining slight penalties from near_backbone, dry5, phobic-fit, sidechain, and hbond_geom_beta_pair. Is it possible to improve these with a new optimization? Is it necessary to improve it? Top score was 210.20. try6-opt2 scored 222.21. The antiparallel sheet at the end of the model is sticking out kind of awkwardly. It's time to think about how to make that part nicer, as the rest of the structure seems to look ok. Sun Jul 9 14:02:45 PDT 2006 Kevin Karplus I'd like to know what the best monomer models are. They can be read from hexamer models: InfilePrefix 6mer/decoys/ ReadConformPDB T0306.try3-opt2.unpack.pdb chain A Sun Jul 9 14:19:40 PDT 2006 Kevin Karplus The gromacs models for the hexamer were not created. I'm making them by running cd 6mer make T0306.mult1 T0306.mult3 T0306.mult4 T0306.mult5 T0306.mult6 (the mult6 target should replace the do6 target when working on a multimer). Sun Jul 9 14:57:21 PDT 2006 Kevin Karplus The 6mer/try6 costfcn likes best the 6mer/try3-opt2 model, but Rosetta likes best 6mer/decoys/T0306.try4-opt2.unpack.gromacs0.repack-nonPC.pdb.gz Sun Jul 9 16:55:34 PDT 2006 Kevin Karplus I'm a little bothered that all the top models seem to be based on the same fold---one that matches the secondary structure prediction rather poorly. Do we have any with more sheet where strands are predicted? Mon Jul 10 12:02:44 PDT 2006 Zack Sanborn Chris and I are running through the various monomeric models, trying to find one that fits the secondary structure predictions better. We know there are a few, but can't remember which. It's a good idea to look back at what we've done anyway: try5-opt2 try6-opt2 Doesn't fit 2nd structure (from alignment) try7-opt2 Ugly breaks, poor beta strands (ProteinShop) try8-opt2 Looks same as try7-opt2 (ProteinShop) try9-opt2 Same as try7/try8 (ProteinShop) try10-opt2 Better looking beta strands, few (if any breaks) Not very protein-like, wedge on wrong side for hexamer model, but does fit 2nd structure pred. (ProteinShop) try11-opt2 Doesn't match 2nd structure, bad breaks. Beta strands peeling apart. This is the model that was made from alignments but using the sheet and helix constraints from try10. try12-opt2 Same as try11. try13-opt2 Same as try12, no obvious breaks, but still doesn't match 2nd structure. try14-opt2 Same as try13. try15-opt2 Same as try14. try16-opt2 Roughly same as try15, but helix was lengthened. However, helix was predicted to be strand. try17-opt2 Same as try16. try18-opt2 A new structure based off an alignment to 2a10 and 2a18, the carboxysome hexamer templates. Structure gets the strands correct and part of helix correct, based off the secondary structure predictions. However, there is a helix where there was predicted to be a loop and a beta-strand. This is the structure that was used to build our hexamers. try19-opt2 Same as try17. try20-opt2 Same as try18. So, we don't REALLY have any models that fit the secondary structure very well, but the try17 family of models might fit the predictions *slightly* better, with more beta-strands. We haven't yet made a hexamer model from this family of models, so we'll do that today (by we, I mean Chris). Mon Jul 10 15:25:34 PDT 2006 Zack Sanborn Well, in the effort to make a few quick, different models, we're restarting the try11 run where we took the sheet/helix constraints that came from the ProteinShop model and made a new structure from the alignments. Well, the model that came from that matched some of what we wanted, but not all of the constraints were satisfied by that structure. The constraints were set fairly low for the try11 run. So, for the new run (try21), I've copied the same script, but ramped up the constraints to 100 (from 10) and weighted the sheet constraints themselves twice the size of the helix constraint. try21 was started on camano at 13:20. Mon Jul 10 15:54:10 PDT 2006 Zack Sanborn In the interest of getting another model out, and with the possibility that try21 might not work, I've started a try22 that takes the try17-opt2 model and really tries to get the same sheet constraints from try21.costfcn forced upon it. try22 was started on camano at 15:50. Mon Jul 10 16:10:46 PDT 2006 Chris Wong We are going to re-polish try10, with some differences. We're going to try to make it better, and not worse. We going to use try10.costfcn. We're going to modify the constraints in the folling way: Template: model3_renumber.pdb... a ProteinShopped model. increase jiggle* and shift* and tweak* ramping up phobic_fit and soft_clashes and breaks and dry_5 removing maybe_metal and maybe_disulfide Our goal is to make something different. This is try23 started on vashon at 1617. Mon Jul 10 16:58:20 PDT 2006 Chris Wong We made some new alignments from 1k2xB and 1seoB. These are the top VAST alignments to our totally aritifical ProteinShop model (each with ~50 residues alignment). We added them as MANUAL_TOP_HITS and made/read the new alignments. We will use the try16.costfcn minus the constraints. This is try24... started on orcas at 1702. Mon Jul 10 18:31:09 PDT 2006 Zack Sanborn Well, all of the trys have finished. We did get some very different looking models. Unfortunately, the models are filled with breaks and show little match to the secondary structure prediction. These models are clearly inferior to the ones made before. This was, in a way, expected given how we were forcing the constraints in two of the models (try21 and try22) and trying an odd template for try24. And, Undertaker couldn't really do anything with the ProteinShop model in try23. The costfcn's with the constraints (try21, try22, try24) like try23 and try10 the best. These models are the ProteinShop models with minor optimizations, and were the models where we got the constraints. SO, this makes sense. However, I hate the way try23 looks. It just looks like it was ProteinShopped. I'm going to make an optimization run for try24 to reduce its breaks and maybe try to pack it a little better. It's the only one out of the new group that has any potential, sad to say. try24 started on camano at 6:50pm. Mon Jul 10 18:57:09 PDT 2006 Zack Sanborn I've looked at the top-2 scoring server models (not including SAM): SPARKS_TS1 and RAPTOR-ACE_TS1. They both are all-beta barrels (?), differing from anything we have since we've been trying to get that predicted helix in our structures. It might be worth starting from one of these models and see where it takes us. It's funny, but the RAPTOR-ACE_TS1 model appears very similar to our ProteinShop model, but has a more convincing twist to the pair of beta-blankets. It's basically what I wanted Undertaker to do to our ProteinShop model, except keep the helix. I haven't looked closely at how well these would align, so it's hard to say if our ProteinShop model would have conformed in such a way. May be simply impossible. However, I know we don't want to be a Meta-server, so I'm not sure if we should go ahead and make a model based on either (or both) of these server models. Tue Jul 11 00:14:54 PDT 2006 Zack Sanborn After getting approval from Kevin to go ahead and polish the server models, I've started two runs that are polishing the SPARKS_TS1 and RAPTOR-ACE_TS1 models. try26 is polishing SPARKS_TS1 and was started on lopez at 0005. try27 is polishing RAPTOR-ACE_TS1 and was started on vashon at 0010. I won't be awake in time to check these models, so hopefully Kevin can take care of it before the submission! I'll send him an email to make sure he knows what's happening. Also, try25 has finished. It scores horribly with its own cost function. But, it is a different looking model. I'm hoping that the server models will be much better. I've updated the superimpose-best.under with the three different models (try20-opt2, try19-opt2, try25-opt2). I'm hoping that the polished server models will provide two good more models. We'll see. Tue Jul 11 07:15:29 PDT 2006 Kevin Karplus I made a secondary.costfcn that is like try1.costfcn, except that it includes only the dssp-ehl2 constraints and not the sheet constraints or rr constraints. It prefers try20, try18, try5, try27, try6, try4, try26, try2, try13, try15 Rosetta's favorites are decoys/T0306.try20-opt2.gromacs0.repack-nonPC.pdb.gz decoys/T0306.try19-opt2.gromacs0.repack-nonPC.pdb.gz decoys/T0306.try27-opt2.gromacs0.repack-nonPC.pdb.gz decoys/T0306.try26-opt2.gromacs0.repack-nonPC.pdb.gz decoys/T0306.try25-opt2.gromacs0.repack-nonPC.pdb.gz decoys/T0306.try23-opt2.gromacs0.repack-nonPC.pdb.gz decoys/T0306.try24-opt2.gromacs0.repack-nonPC.pdb.gz but some of the early models have not had the gromacs0.repack-nonPC run. Redoing try5, try6, try4, try2, try13, and try15, does insert some more in the top list. decoys/T0306.try20-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try13-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try19-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try15-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try27-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try5-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try2-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try26-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try25-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try4-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try6-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try19-opt2.repack-nonPC.pdb decoys/T0306.try23-opt2.gromacs0.repack-nonPC.pdb decoys/T0306.try24-opt2.gromacs0.repack-nonPC.pdb So what about try13, try19, try15, try17, try5? Notes above say try13 ~ try12 ~ try14 ~ try15 ~ try16 ~ try17 ~ try19 Why are we prefering try19 over try13? Well, it scores better with a constraint-less costfcn (try19), though the secondary.costfcn prefers try13. I like try13 better myself, so will choose it over try19. try5 is similar to try20, but I prefer try20. I'd rather include an ugly ProteinShop model (try23) than try25, even Rosetta likes try25. So the submission will be try20-opt2 try13-opt2 try23-opt2 # from ProteinShop try27-opt2 # from RAPTOR-ACE_TS1 try26-opt2 # from SPARKS_TS1 Tue Jul 11 08:00:05 PDT 2006 Kevin Karplus So submitted.