Wed Jun 30 08:30:18 PDT 2004 T0222 Due 18 Aug 2004 Wed Jun 30 09:27:02 PDT 2004 Kevin Karplus T0222 looks like a comparative model with c.1.10.4 as the SCOP domain. Another TIM barrel. Sat Jul 17 21:31:05 PDT 2004 Kevin Karplus The active site probably consists of conserved polar groups on the c-terminal end of the barrel and on the loops: H236, E234, S212, H214, D247, C181, Y185, George's contact predictions give us others in this network: S130, T108, H180, but these predictions re pretty weak. We have to decide whether we have all the strands correctly chosen, or whether there is an earlier insertion, moving the identities of the strands. Sat Aug 7 19:29:46 PDT 2004 Kevin Karplus George is taking over this target. I'll set up and run try2 for him, since he's out of town until tomorrow. I'll do a "make -k" to make sure that that t04.many.frag file is created, "make extra_alignments" to make alignments for all the MANUAL_TOP_HITS, and "make all-align.*" to update the all-align.a2m.gz file. The try2.under file will read both t2k.many.frag and t04.many.frag, so will still output Template.atoms. For try3, George will have to comment out that line. The try2.costfcn includes helix and strand constraints and rr.constraints, but not any sheet constraints---that will have to wait until Geoge takes over the target and tries to adjust the barrel (if adjustment is needed). Sun Aug 8 10:29:10 PDT 2004 Kevin Karplus The try2 model is awful---strands and helices are badly misaligned. We will definitely need sheet constraints to build the correct alignment. George says he knows how to build TIM barrels now, so this will be a good exercise for him. Sun Aug 8 20:06 2004 Bret Barnes I ran a VAST alignment on try2, but it hasn't finished yet. ID: VS60580 Password: bb7509bb Sun Aug 8 23:34:05 PDT 2004 George Shackelford We've already got a good handle on this structure, I don't know if we need a VAST alignment. I am setting up the label of strands and mark the current and likely strands for the barrel. I am going to primarily work with t04. I haven't seen anything to indicate either t2k or t04 is better so I'll default to t04 (I just trust it more). Ok, I'll need to finish filling this out tomorrow. t04 strands (and helices): StrandConstraint A3 I6 0.7514 # s1 StrandConstraint V17 E22 0.7023 # s2 HelixConstraint L31 E43 0.9221 HelixConstraint L80 Q82 0.6411 HelixConstraint A88 E100 0.8452 StrandConstraint I104 S107 0.6812 # s3 HelixConstraint E112 S120 0.6339 StrandConstraint A125 I128 0.6263 # s4 HelixConstraint P137 R144 0.8686 StrandConstraint M149 S152 0.6458 # s5 HelixConstraint I158 A170 0.9152 StrandConstraint I176 H180 0.675 # s6 HelixConstraint L194 A203 0.6472 StrandConstraint V208 G210 0.6047 # s7 HelixConstraint P222 R227 0.7274 StrandConstraint K231 K235 0.619 # s8 HelixConstraint P254 K272 0.7183 HelixConstraint V279 L283 0.6495 HelixConstraint I294 F301 0.6507 StrandConstraint I306 T309 0.6939 # s9 StrandConstraint I323 L326 0.6112 # s10 HelixConstraint P337 T344 0.8096 StrandConstraint R348 V350 0.6141 # s11 HelixConstraint W361 I364 0.6184 try1 sheets: SheetConstraint F4 I6 V11 K9 hbond F4 # s1a,s1b SheetConstraint F18 A21 A230 I233 hbond I20 # s2||s8 # we need a way to connect s2,s2c (where s2c is really SheetConstraint L66 T69 I104 S107 hbond Y67 # s2c||s3 SheetConstraint F105 T108 S124 K127 hbond S107 # s3||s4 SheetConstraint A125 I128 P148 F151 hbond F126 # s4||s5 SheetConstraint P148 T153 Q175 H180 hbond M149 # s5||s6 SheetConstraint A177 H180 E206 I209 hbond I178 # s6||s7 SheetConstraint V208 G210 A230 L232 hbond I209 # s7||s8 # SheetConstraint A251 L252 E234 K235 hbond L252 # s8b||s8 I notice that the SheetConstraints contain Strands not indicated in the main list. The barrel helices are facing the wrong way in several cases(??). Wed Aug 11 12:30:03 PDT 2004 George Shackelford Ok, let's see if we can fix up the sheets a bit. Is the short and vague 66-69 strand part of the barrel or is it an insert? If we take it out, we still have other possiblities. s2,s3,s4,s5,s6,s7,s8 look solid - but is 9,10,11? or s2[bc] the missing link? t2k.ehl2 hints strongly at s2b. In fact, the concensus of t2k,t04 ehl2,str2 is that s2b is the only possibility. I'll work with that... I put the try1 sheet constraints in and fix the up for using s2b. I also included the t04.ehl2 constraints. TRY3 running on crow. Wed Aug 11 23:00:00 PDT 2004 George Shackelford The new strand, s2b, misfired. I'm going to give it a shift of one residue while aligning it to the s2 and s3. TRY4 started on caw. Thu Aug 12 12:39:03 PDT 2004 George Shackelford Try4 is looking good (the barrel seems in place) but some of the strands have polars on the wrong side. I need to flip those strands and that can be difficult. I'm doing a VAST search to get a better alignment: password: wn34pp7mv VAST id: VS60729 I may still try and flip them to see what happens. Thu Aug 12 17:24:22 PDT 2004 The results are back from VAST and I selected several well scoring sequences, got the alignment and pasted it into a file called T0222.VAST.a2m. I just need to make the editing changes. I noticed that the second best scoring p-value matches the second best scoring of the a2m alignments. I wonder if that helps. Well on to editing. Fri Aug 13 11:14:53 PDT 2004 George Shackelford First I set up the current alignment: 1234567890 FiIA (18-21,19) R VKfQM (49-53,51) R IFLsT (104-108,107) R SAfKI (124-128,126) PmIFST (148-153,149) QIAiMH (175-180,178) EAViG (206-210,209) xAkLI (230-233,231) FiIA (18-21) I've just realized I some of the hbonds wrong! No wonder this thing is so flipped around! Furthermore the full shift is only 6 rather than 8. VFIiAE (17-22,20) VKFqM (49-53,52) IFlST (104-108,107) R SAfKI (124-128,126) PmIFST (148-153,149) QIAiMH (175-180,178) EAViG (206-210,209) KLiEK (231-235,231) VfIIA (17-21) I also realized that the problem with the reversals is in the area that I did manually to get the s2b in place. I must have caused the problem; I probably can fix it manually without having to do the VAST trick and the special alignment. Running TRY5 on peep. Fri Aug 13 23:12:58 PDT 2004 George Shackelford Try5 is better but there are still a few misfits. More tomorrow. Sat Aug 14 10:12:58 PDT 2004 George Shackelford I need to realign s2b up a residue so 51 or 49 is the hbond. 50 needs to point in. I also need to shift s3 up one. I might later change s6, but it's ok for now. New alignment: VFIiAE (17-22,20) AVKfQ (48-52,49) changed IFlST (104-108,106) changed hbond AfKIA (125-129,126) changed PmIFST (148-153,149) QIAiMH (175-180,178) EAViG (206-210,209) KLiEK (231-235,231) VfIIA (17-21) I have a new try6 running on abyss. I can't seem to reach peep. Sat Aug 14 20:23:12 PDT 2004 Kevin Karplus In try6, the follow residues appear to be on the wrong side of the sheet K50 S107 (but T018 on the right side, so hard to tell) T153 (but S152 on the right side, so hard to tell) H180 (but conserved C181 on right side, and conserved I178 on right side, so probalby correct) It may be past time to stop dinking around with the constraints and start from an alignment that will give us the barrel we want. I thought that George was going to edit an alignment yesterday, but I see no evidence that he has done so. The 1d9eA-T0222-try4.vast.a2m alignment is still around for editing. I don't understand George's notation above for the barrel that he saw in try4-opt2. Reading the sheet constraints and looking in rasmol at try4-opt2, I see the following barrel: 17: VFIIAEag 48: aVKFQMfq 103: vIFLSTvc 124: sAFKIas 148: pMIFSta 174: nqIAIMHc 206: eAVIGFsd 230: aKLIEKhft 17: VFIIAEag This has a proper slip of 8, but some strands are flipped. Here is a possible correction: polar: * * * * * * * 17: VFIIAEag 48: aVKFQMfq back one 103: vIFLSTvc forward one 124: sAFKIas 148: pMIFSta 174: nqIAIMHc forward one 206: eAVIGFsd forward one 230: aKLIEKhft 17: VFIIAEag polar: * * * * * * * SheetConstraint I19 E22 V49 Q52 hbond A21 SheetConstraint F51 F54 I104 S107 hbond F51 SheetConstraint F105 T108 A125 I128 hbond L106 SheetConstraint A125 A129 P148 S152 hbond F126 SheetConstraint M149 A154 Q175 H180 hbond M149 SheetConstraint I176 H180 A207 F211 hbond A177 SheetConstraint V208 S212 K231 K235 hbond V208 SheetConstraint K231 I233 V17 I19 hbond K231 OK, I think I've edited 1d9eA/1d9eA-T0222-kevin1.a2m to meet make the desired alignment. Now I'll set up try7 to use the constraints and the alignment to make a barrel (I hope). Sun Aug 15 08:56:19 PDT 2004 Kevin Karplus Did I goof on the constraints or the alignment? For try7, the strand with Q175 is flipped and I178 is flipped. Looking back, that is what I requested---probably a bad request. Should I fix the alignment? 148: pMIFSta 174: nqIAIMHc backward one 206: eAVIGFsd backward one 230: aKLIEKhft SheetConstraint M149 T153 I176 H180 hbond M149 SheetConstraint I176 H180 A207 F211 hbond I176 SheetConstraint V208 S212 A230 E234 hbond I209 I'll start try8 from 1d9eA-T0222-kevin2.a2m and see how well it works. Currently try5-opt2.repack-nonPC is best with an unconstrained cost function. Sun Aug 15 10:21:28 PDT 2004 George Shackelford I'm going back to try6 and make adjustments that Kevin noticed needed to be made but I am going to stick to my approach of just adjusting the constraints. As long as those changes work, I don't think I'll try the modified alignment. The notation I used was: XXXxXX ( - ,) where the X's represent the strand and the lower case x represents the hbond. The use of the visual hbond in the strand helps me to be sure that the hbonds are lined up correctly (otherwise the barrel breaks). I'm going to start a try9 (before Kevin gets the number). First, I note that he has the correct value for the s2 strand's hbond (A21). Second, I like his strand for s2b but shifted forward by 2. [NOTE: s2b is StrandConstraint 48-53. It is not added to the regular constraints because it is already defined in the SheetConstraints and is quite stable] current: VFIiAE (17-22,20) AVKfQ (48-52,49) IFlST (104-108,106) AfKIA (125-129,126) PmIFST (148-153,149) QIAiMH (175-180,178) EAViG (206-210,209) KLiEK (231-235,231) VfIiAE (17-22) new: FIIaEA (18-23,21) changed hbond and shifted AVKfQ (48-52,49) IFlST (104-108,106) AfKIA (125-129,126) PmIFST (148-153,149) QIAiMH (175-180,178) EAViG (206-210,209) KLiEK (231-235,231) FIIaEA (18-23) SheetConstraint F18 A21 L232 K235 hbond I233 10.0 # s2||s8 changed SheetConstraint I19 E22 A48 F51 hbond I21 10.0 # s2||s2b changed SheetConstraint V49 Q52 I104 S107 hbond F51 10.0 # s2b||s3 SheetConstraint F105 T108 A125 I128 hbond L106 10.0 # s3||s4 SheetConstraint A125 A129 P148 S152 hbond F126 10.0 # s4||s5 SheetConstraint P148 T153 Q175 H180 hbond M149 10.0 # s5||s6 SheetConstraint A177 H180 E206 I209 hbond I178 10.0 # s6||s7 SheetConstraint A207 G210 K231 E234 hbond I209 10.0 # s7||s8 Sun Aug 15 14:54:06 PDT 2004 Kevin Karplus In try8, the strands seem to be basically correct, except for E22, but that is on the right side---it just has a funny rotamer. I'm not sure why George wants to change the alignment of A48-M53---I don't see how sliding it forward by two would help. Can you explain, George? Our big problem now is not the barrel, but the stuff around it. George's try9 cost function likes try5, as does the unconstrained costfcn. I'll try to polish up try8, since George has no interest in it. Sun Aug 15 15:15:14 2004 George Shackelford The reason for shifting A48-M53 (a.k.a. strand s2b) is because it was linking to I104-S107 by only a couple of residues. I thought that it might fit better; just a guess. If the results of try9 look good, then I'm going to start tying parts of the remaining structure down (that part > residue 233). I have a number of rr constraints in that area as well as conserved residues to guide me. The liking of try5 may have more to do with how the sequence above 233 worked out rather than the smaller adjustments made to the barrel. Sun Aug 15 19:00:34 PDT 2004 Kevin Karplus try9 finished, but the files were called try6... since George had not edited try9.under to replace try6 by try9. I'll rename the files in decoys/ (luckily the REAL T0222.try6* files were gzipped, so didn't get stepped on, and I had backup on my laptop in any case). The try9 run was optimizing for try6.costfcn, so may not have been very useful (depending on whether the try6.costfcn was useful). try9-opt2 scores very slightly better than try6-opt2 with the try9 costfcn, but not as well as try5-opt2. The try10.costfcn currently scores try10-opt1 best, slighlty better than try8-opt2. The unconstrained costfcn still scores try5-opt2.repack-nonPC best. Try10-opt1 has moved a little up fro try8-opt2, but try6, try9, try3, try2, and try1 all still beat it. The biggest problems seem to be soft_clashes and backbone_clashes, though the pred_alpha(2k and 04) values also need help. try5-opt2 forms more helices which helps it somewhat. I'm going to start a subdomain prediction for H236-end. Sun Aug 15 19:45:53 PDT 2004 Kevin Karplus The 236-end/ setup run has not finished yet, but I'm seeing pretty clear signals that the domain is a b.85.1.1 domain. If we can fold it neatly in just one or two tries, then we can paste it into the whole-chain prediction and reoptimize (making sure to include alignments and fragments from the 236-end/ directory). Sun Aug 15 20:11:42 2004 George Shackelford I had made some notes which I tried to copy and paste here (since Kevin and I are trying to work in the same README) but managed to lose them all. Basically I found the same issue about failing to change try6's to try9's. Usually I use my simple utility, 'newtry,' but couldn't this time. Need to fix that. Well I don't know what to do since Kevin is ahead of me by using the 236-end to spot a subdomain. Seems he has everything going. Wish I could help. I was going to do a try using the rr constraints to see if they would fit stuff together in 236-end... Sun Aug 15 20:50:43 PDT 2004 Kevin Karplus 236-end seems to be working fairly well. There is a fairly strong fold-recognition hit (b.85.1.1), agreed on by both the t2k and t04 methods, with 1ucsA as the top representative. Either tonight or tomorrow, I'll start a 236-end/try2 run with t04 favored over t2k. If that works, then tomorrow I'll do a final polishing run on 236-end, then cut-and-paste the domain with whatever our best model of the barrel is. It might be worthwhile to build a 1-240 subdomain of just the barrel, seeding the decoys file with a truncated version of our best barrels. That way we could be polishing the barrel without the noise of the unfolded C-terminal domain. I added the 236-end hits to the MANUAL_TOP_HITS list for this directory and made extra_alignments and remade all-align.a2m.gz From karplus@soe.ucsc.edu Sun Aug 15 21:00:17 2004 Date: Sun, 15 Aug 2004 21:00:15 -0700 From: Kevin Karplus To: ggshack@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: T0222 There is something you could do in parallel with my work on T0222/236-end. You could set up a subdomain 1-240, and get it running. (You might want to grab the initial files for .under from casp6/starter-directory, to be sure of getting recent ones---the XXX0000 in them needs to be replaced by T0222.) If you seed the decoys directory with truncated versions of our best barrels, and use the best costfunctions, then you should be able to do some polishing work on the barrels without the nose of the unfolded C-terminal domain. Tomorrow afternoon or evening, we can paste the two domains back together and do one final polish on the whole thing. ============================================================ Sun Aug 15 21:08:12 PDT 2004 Kevin Karplus try10-opt2 scores best with the try10 costfcn, but has too many clashes to do well in the unconstrained costfcn. It needs some polishing. From karplus@soe.ucsc.edu Sun Aug 15 21:41:13 2004 Date: Sun, 15 Aug 2004 21:41:12 -0700 From: Kevin Karplus To: ggshack@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: domain boundary in T0222 You may want to set the domain boundary a bit later for the barrel--around G274. There is some evidence in the 236-end subdomain that there is an initial helix that is not part of the main prediction---that could well be a final helix of the barrel or a linker. From ggshack@pacbell.net Sun Aug 15 21:57:27 2004 MIME-Version: 1.0 From: George Shackelford To: Kevin Karplus Subject: Re: domain boundary in T0222 Date: Sun, 15 Aug 2004 21:57:24 -0700 In-Reply-To: <200408160441.i7G4fCcJ028473@cheep.cse.ucsc.edu> Content-Disposition: inline X-Spam-Checker-Version: SpamAssassin 2.64 (2004-01-11) on fs.cse.ucsc.edu X-Spam-Level: X-Spam-Status: No, hits=0.0 required=3.0 tests=none autolearn=no version=2.64 Ok, I was just getting started so I'll make the change. We can always trim for the best match. - George On Sunday 15 August 2004 09:41 pm, you wrote: > You may want to set the domain boundary a bit later for the > barrel--around G274. There is some evidence in the 236-end subdomain > that there is an initial helix that is not part of the main > prediction---that could well be a final helix of the barrel or a linker. Sun Aug 15 22:25:25 2004 George Shackelford make of the domain 1-274 started on ribbit. Sun Aug 15 22:58:30 PDT 2004 Kevin Karplus The 236-end/ try2 run did worse (on the try2 cost function) than try1. I guess it is time to try polishing with the try2 cost function, then again with an unconstrained cost function. Mon Aug 16 07:10:50 PDT 2004 Kevin Karplus I'm downloading the 1-274 directory to my laptop, but do not expect to work on it myself. George ran the 1-274 run with the default try1.costfcn. He should now copy some of our best models (try5-opt2, try6-opt2, try9-opt2, try10-opt2, try3-opt2, try8-opt2) into 1-274/decoys and truncate them (maybe giving them names like "from-whole-try5", to distinguish them from the runs done in 1-274 itself. He should also copy some of our best conjectures about constraints (try10.costfcn and whatever his current favorite is) as cost functions, again removing references past 274. He can do a run or two from alignments with these cost functions (perhaps a couple of runs in parallel to save time), then do a polishing run using the best cost function and all the models. After that, one more polishing run using an unconstrained cost function would probably be called for. For 236-end/, I'm a bit worried that the current model has stuff wrapped around the initial helix. If I'm right, and that helix belongs to the barrel domain, then the stuff I'm packing on the buried side of it all wrong. Perhaps I should start yet another subdomain, 273-end, starting at the end of the helix with Q273. Mon Aug 16 09:29:44 2004 George Shackelford I'm looking at Kevin's notes and the 1-274/try1. He's right. Need to adjust constraints and get some truncated pdb's (esp the good scoring try5). I am a bit worried about the extra loops off of the barrel; they probably interact with his domain and we'll have to take them into account when putting things back together. I've added truncated versions of try5,try6,try9. May add others later. I've put together a list of constraints for 1-274/try2 and will start it on peep. Mon Aug 16 15:10:44 PDT 2004 George Shackelford The results of the try2 is out and it looks real nice, except that I need to turn 103-108 around. A shift up of one and a twist would work and might help the helix reach the next strand correctly. I could also tighten the break and adjust other values. Otherwise we're finished here. I need to look at the loops on top to see where to tuck them. There may be a flap at the top as well. Hmm. 103-108 should be right, but it's not. 106 seems to miss it's match. Odd. 48-53 also seems to be wrong though it has the right match. This is what I had for myself earlier: FIIaEA (18-23,21) AVKfQ (48-52,49) IFlST (104-108,106) AfKIA (125-129,126) PmIFST (148-153,149) QIAiMH (175-180,178) EAViG (206-210,209) KLiEK (231-235,231) FIIaEA (18-23) SheetConstraint F18 A21 L232 K235 hbond I233 10.0 # s2||s8 SheetConstraint I19 E22 A48 F51 hbond I21 10.0 # s2||s2b SheetConstraint V49 Q52 I104 S107 hbond F51 10.0 # s2b||s3 SheetConstraint F105 T108 A125 I128 hbond L106 10.0 # s3||s4 SheetConstraint A125 A129 P148 S152 hbond F126 10.0 # s4||s5 SheetConstraint P148 T153 Q175 H180 hbond M149 10.0 # s5||s6 SheetConstraint A177 H180 E206 I209 hbond I178 10.0 # s6||s7 SheetConstraint A207 G210 K231 E234 hbond I209 10.0 # s7||s8 I like them. I'm going to use them for the next try. I have some loops I'd like to take care of: 54-87, 238-250, 181-190. These may or may not have something to do with 273-end. Mon Aug 16 17:04:28 PDT 2004 Kevin Karplus The 273-end/ try1 run was terrible, probably because the t04 secondary structure predictions were very weak and inconsistent with the stronger t2k predictions. The top few alignments agreed with the older prediction of that domain in 236-end/ try3-opt2, so I set up a try2 run in 273-end with more appropriate constraints and am running try2. Assuming that works ok, I'll do a try3 polishing run and be ready to merge with George's barrel predictions. Mon Aug 16 19:48:14 2004 George Shackelford 1-274/try3 is done, looks good, but scores a bit worse than try1 and try2. I think I'll go for a final polishing run prior to merging. I'm cranking up the wet and dry, removing the constraint stuff and option, commenting out the TryAlls and including the read-pdb.under. I think that about covers it. Try4 on crow. Mon Aug 16 20:01:52 PDT 2004 Kevin Karplus 273-end/ try2-opt2 looks pretty good. I'll do a polishing run without constraints (try3, running on cheep). From karplus@soe.ucsc.edu Mon Aug 16 20:26:00 2004 Date: Mon, 16 Aug 2004 20:25:59 -0700 From: Kevin Karplus To: ggshack@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: T0222/1-274 Are you sure that try4.under does what you want? Since try4.costfcn scores try1-opt2 best (by quite a bit), you will be mainly polishing that model. I suggest you turn the dry weights down to more reasonable numbers (like no more than 20). Tweak the weights until the model you want to optimize barely scores the best. ============================================================ Mon Aug 16 21:15:14 PDT 2004 Kevin Karplus I think that 273-end is done. The try3-opt2 model looks pretty good to me. Hooking up to the barrel may be a bit tough---we may have to place the second domain roughly near where it goes (not overlapping the barrel) and let undertaker optimization try to place it. Mon Aug 16 21:48:47 PDT 2004 Kevin Karplus Since George has not responded to my query about try4, I'm going to start a try5 that uses t04.dssp-ehl2.constraints and sheets from try10-opt2 in the whole model. George may disagree with this barrel, and we may end up submitting two different barrel domains. George has responded, and try4 is doing what he wants. I'll go ahead with a try5 anyway, to see what comes out. The 1-274/try5.costfcn likes try1-opt2 best, better than from-whole-try10-opt2, so maybe 1-274/try1-opt2 is better than I realize. I'm still curious what comes out from try5, but I'll probably have to wait until the morning. Tue Aug 17 07:41:27 PDT 2004 Kevin Karplus Well, 1-274/ try5-opt2 seems to be a bit of a bust---it doesn't do as well as any of try1, try2, try3, or try4, even on the try5 costfcn. George's try4-opt2 will most likely be the barrel we submit, unless George has another one he likes better. I'm really convinced, though that K50 and Q52 face the wrong way. H180 may also, though the orientation of I178 seems to indicate that that strand is correctly oriented. I want to push K50 up to align with E22 and Q52 to align with T108. Maybe I can make a chimera of try4 and from-whole-try10 that fixes just that strand? I could copy G45-Q55 after superimposing so that E22 lines up well. Despite some bad breaks and clashes from the cut-and-paste job, try4-try10-chimera scores nearly as well as try1 and try4 with both the try4 and the try5 cost functions. I'm going to try polishing it up and see if it gets to be any good. Tue Aug 17 10:46:19 PDT 2004 Kevin Karplus 1-274/ try6-opt1 beats try4-opt1 (though not quite try4-opt2) on the try6 unconstrained costfcn. I think that try6-opt2 will be the model to use. One minor problem with try6-opt1---I don't like the rotamer for E22---perhaps I should add a distance constraint for E22.CD and K127.NZ or Q52.NE2 (maybe H236.NE2 also). Tue Aug 17 11:22:10 PDT 2004 George Shackelford Ok, I'm convinced about try6. Tue Aug 17 11:22:10 PDT 2004 Kevin Karplus While 1-274/ try7 tries to clean up E22 in try6, I'm going to try to put the domains together. I can always superimpose try7 on the result and do cut-and-paste to pick up any improvements try7 makes. Tue Aug 17 12:00:08 PDT 2004 Kevin Karplus try11 has been started to try to polish the two-domain.pdb model. It includes some constraints, so I may want to do another polishing run when it is finished (perhaps after cutting-and-pasting in the 1-274/ try7-opt2. Tue Aug 17 12:15:35 PDT 2004 Kevin Karplus Well, the 1-274/ try7-opt1 run has gone in the wrong direction, pulling opt4-try2 out of the pool, rather than fixing try6-opt2. I think I'll kill try7 and start over on try8, with just try6 models as input. Tue Aug 17 12:27:21 PDT 2004 Kevin Karplus The try11 run seems to be removing clashes and breaks, but I don't know if it is destroying the domains in the process. It'll be about an hour befor the try11-opt1 results are available to look at. If the domains look damaged, I could try using try11-opt1 as a framework for re-superimposing the individual domains, and start over. I hope it doesn't come to that! Tue Aug 17 13:58:12 PDT 2004 Kevin Karplus The 1-274/ try8 run seems to be doing what I wanted, pulling E22 out of its strange conformation. The try11 run in the main directory seems to be having some trouble---it unwound a bit of the last helix of the barrel to attach to the second domain. If it doesn't fix this problem by try11-opt2, then I may have to do another cut-and-paste to fix it manually. I don't think the second domain is docking properly against the first, but there probably won't be time to explore that. Tue Aug 17 14:44:26 PDT 2004 Kevin Karplus The 1-274/ try8 run looks pretty good and scores best with an unconstrained costfcn. There is a hydrophobic patch around A202 and P198 where the second domain could dock, and the polars in the middle look pretty reasonable. As soon as try11 finishes, I'll do a cut-and-paste to put in the try8 barrel. Then I'll do another polishing run, perhaps with some constraints to try to pack the second domain against A202, F204, P205, W165? Another strong possiblity is that the second domain backs against the bas of the barrel (A2, M1, A230). Maybe it would be best if I didn't try to dock the second domain, as I really have not idea where it would fit! Tue Aug 17 15:44:36 PDT 2004 Kevin Karplus try11-opt2 did not fix the broken helix, so I'll patch in the two domains again: ReadConformPDB 1-274/decoys/T0222.try8-opt2.pdb # domain 1 ReadConformPDB 273-end/decoys/T0222.try3-opt2.pdb # domain 2 superimposing them on try11-opt2. The resulting chimera is decoys/two-domain-11.pdb. It scores almost as weel as try11-opt2 with an unconstrained costfcn, despite the large break. I'll try a quicker polishing run on it, with an unconstrained costfcn. Tue Aug 17 16:39:01 PDT 2004 Kevin Karplus The try12-opt1 model has closed the gap reasonably, without destroying either domain. I don't think the domains are properly packed against each other, but that docking may be tough. Maybe I should start a parallel run with constraints for F240-A202, A324-M199 F301-P205 to try to fold the hydrophobic hole in the second domain over the hydrophobic hump on the first domain. I'll run that as try13, but I'll submit try12-opt2 as soon as it is ready. I plan to submit try12-opt2.pdb # best unconstrained cost try13-opt2.pdb # different packing of domains try4-opt2.repack-nonPC.pdb # best rosetta energy (yuck) try1-opt2.pdb # full auto T0222-1o60A-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 # top template Since try13-opt2 is not done yet, I'll temporarily include two-domain-11.pdb in its place. Actually, depending how things look, I may prefer to keep two-domain-11 in place of one of try12-opt2 or try13-opt2, since it has the domains separately solved without distortions to make them join. Tue Aug 17 18:01:22 PDT 2004 Kevin Karplus Foo! try12-opt2 moved the E22 back to the wrong side of the sheet! I hope that try13-opt2 keeps it in the right place. try12-opt2.repack-nonPC is now the first-place Rosetta energy. try13-opt1 does not look promising. It has pulle the last helix off the barrel AND tucked E22 on the wrong side of the barrel. The submission try12-opt2.pdb # best unconstrained cost try13-opt2.pdb # different packing of domains try12-opt2.repack-nonPC.pdb # best rosetta energy try1-opt2.pdb # full auto T0222-1o60A-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 # top template will probably be final. Tue Aug 17 21:01:22 PDT 2004 Kevin Karplus try13-opt2 is indeed terrible---a helix was ripped of the barrel rather than repositioning the 2nd domain. Fri Sep 24 20:55:56 PDT 2004 Kevin Karplus Our best model was try5-opt1 name length missing_atoms rmsd rmsd_ca GDT smooth_GDT T0222.try5-opt1.pdb.gz 373 0.0000 15.5795 15.2065 -43.7849 -41.3194 model1.ts-submitted 373 0.0000 17.5195 17.1746 -41.0615 -38.0673 try12 model3.ts-submitted 373 0.0000 17.5835 17.1746 -41.0615 -38.0703 try12 repacked model2.ts-submitted 373 0.0000 17.7351 17.4048 -40.7123 -37.7921 two-domain-11.pdb robetta-model1.pdb.gz 373 0.0000 20.7591 20.4069 -37.9888 -35.7288 robetta-model3.pdb.gz 373 0.0000 15.7328 15.4223 -37.5698 -35.4240 model5.ts-submitted 373 943 8.6635 8.0095 -37.6397 -34.2065 alignment robetta-model4.pdb.gz 373 0.0000 15.5146 15.1354 -35.0559 -33.0092 model4.ts-submitted 373 0.0000 23.2316 22.6429 -35.6844 -32.9217 full auto robetta-model5.pdb.gz 373 0.0000 24.3329 23.9662 -33.9385 -31.9215 robetta-model2.pdb.gz 373 0.0000 16.1536 15.6608 -33.8687 -31.8432 The Rosetta repacking that took model 1 to model3 did not reduce the all-atom rmsd. We did not submit our best model, but our model 1 was the best model we submitted, beating the full auto and the alignment. We may have to re-evaluate after splitting into domains.