Tue Jun 15 08:31:51 PDT 2004 T0204 DUE 5 July 2004 Tue Jun 15 09:38:55 PDT 2004 Kevin Karplus Make killed on abyss---it has the broke version of grep that takes forever---DON'T run target99 or target2k on abyss! (It also is missing the perl modules needed for a2m2html.) Tue Jun 15 10:03:23 PDT 2004 Kevin Karplus This looks like a close comparative model, with 1hxpA, 1gupA, and 1guqA as the best templates. We should probably modify the 'try*.under' scripts to limit the templates to d.13.1 superfamily: 1hxpA 1gupA 1guqA 1emsA 1kpeA 1kpf 1fit 2fhi. Tue Jun 15 16:19:30 PDT 2004 Kevin Karplus Try1 hasn't finished yet (the problem with running two targets at once on the same machine!), but I can already see that there is a disulfide (or, more likely, metal-binding site): C63-C66 (maybe C73), C217-C217-C219 I wonder if C331 is part of the C62-C66 cluster---that would require packing the long linker in a very different way. More likely the H134, H184, H186 are involved. with C62 and C66. I'll have to look at the templates and see whether they have metal-binding sites. Tue Jun 15 16:58:25 PDT 2004 Kevin Karplus The main work needed on this one will be the linker: approx R41-W87. From karplus@soe.ucsc.edu Tue Jun 15 17:38:50 2004 Date: Tue, 15 Jun 2004 17:38:48 -0700 From: Kevin Karplus To: sol@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: T0204 Lots to see in T0204. 1) when you make the next "try*.under" script, include the read-alignment.under files to get variant alignments. 2) Look at the 1hxpA rasmol file. In particular, look at select ligand or cys or his (then spacefill) There are metal-binding sites that we probably want to preserve---make constraints to hold the active parts of the residues (the nitrogens and sulfurs) at about the right distances from each other. -------------------------------------------------- Thu Jun 17 16:30:27 PDT 2004 Sol Katzman Looking at the Zn and Fe binding sites from 1guq and 1hxp shows the residues (identical in both of these templates): binding Zn: C52 C55 H115 H164 (from loops) binding Fe: E182(from helix) H281 H296 H298(from strands) For the Zn binding, in T0204 the cysteines are likely C63 C66 and on the loop-equivalents to 1guq, there is H133, H184 For the Fe binding, in T0204 the histidines are likely H295,H310,Q314? and from the helix-equivalent to 1guq, there is D204 (or D208?) which are currently on the wrong side of the helix. For try2: Set constraints to pull these binding sites together. Copied the strand and helix constraints that all look okay from T0204.t2k.dssp-ehl2.constraints Included the read-alignment.under files for the homologs. Minor oops. In try2.under I did not have "OutFilePrefix decoys/" so I manually moved the output files to the decoys directory and (in violation of protocol) post-modified the try2.under file. Fri Jun 18 21:46:09 PDT 2004 Sol Katzman The putative Zn binding site now has C63,C66,H133,H184 all close together but it had to break the chain to get the the 2 Hs to come in. The Fe binding site seems mostly unchanged from try1. It may take something more radical to get an acid like D204 from the other side of the helix. The H310 seems to have gotten a rotamer that brought its ring closer to H295, but there was not much change in H295 or Q314. Sat Jun 19 08:16:54 PDT 2004 Kevin Karplus I made a new "define-bond" rasmol script for Sol's binding sites. It doesn't DO anything, just defines sets bindzn, bindfe, and bind. I'm not entirely convinced that D204 and D208 belong to the active site. I would expect high conservation at a binding site, so H184 and H186 clearly important. Why was 186 excluded from bindzn? unless it is part of an adjacent site for H186, Q188, and N173. N94 and Y96 are also conserved in this 3D neighborhood. We have a cluster of polar conserved residues: C66, N94, Y96, H184, H186, Q188, N173, S181 I'd be tempted to try to cluster the active ends of all of these! I'm less sure about putting in the C63 and H133, though they plausible metal-binding residues in the right neighborhood. The lack of conservation makes me slightly less fond of them. Of course, I've not looked at the original template to see what the binding residues are there---the conserved polars residues may be playing some othe role. At the other binding site, H310 and Q314 are conserved, and H295 seems a likely partner, but I'm not so fond of D204 and D208. Y209 seems like a good possibility, and maybe C216, C217, and C219 I've defined these potential clusters in "define-bind". I'll leave it up to Sol to add constraints to hold their active atoms near each other. Mon Jun 21 13:14:50 PDT 2004 Sol Katzman For the Fe binding site, E182 (*) in 1guqA participates, coming out of the helix that is pairwise aligned to T0204 P197-T213 (the numbering for 1guqA seems to be off by one): 200 210 220 | | | T0204 LPVVPPTVSSRLDGTKDYFEETGKCCLCE 1guqA NSFLPNEAEREDRLQKEYFAEQKSPMLVD | |* | | 170 180 190 200 And the 3 Histidines (*) from 1guqA are H281,H296,H298 290 300 310 | | | T0204 KQLN-DPPYNYMIHTSPLKvtESQLPYTHWFLQIVPQ 1guqA NLFQCSFPYSMGWHGAPFN..GEENQHWQLHAHFYPP | * | * * | 270 280 290 300 That was the justification for try2. Nevertheless, for try3 I am using Kevin's suggested set for the Fe binding site. I am leaving the Zn binding site unchanged, and am increasing the weight for 'break' to 30. -------------------------------------------------------------------------- These are the two SCOP families, and the 3 high percent-identity alignments T0204 is much closer to d.13.1.2 than to d.13.1.1 (which really corresponds to a C-terminal subset of d.13.1.2): SCOP d.13.1.2 Hexose/Galactose/Glucose 1-Phosphate Uridyltransferase 1hxqA,B 11/1997 E.coli Fe,Zn 1hxpA 11/1996 E.coli Fe,Zn 1guqA 11/1997 E.coli Fe,Zn,K 1gupA 10/1996 E.coli Fe,Zn,K SCOP d.13.1.1 Protein Kinase Inhibitor/Histidine Triad Nucleotide Binding Protein 1kpf 03/1998 Human 1rzyA 03/2004 Rabbit 4rhn 06/1997 Rabbit SCOP d.13.1.1 Fragile Histidine Protein 1emsA 03/2000 Worm C-term domain 281-440 1fit 11/1997 Human 2fhi 06/1998 Human -------------------------------------------------------------------------- Tue Jun 22 10:41:13 PDT 2004 Sol Katzman The try3-opt2 result still does not look very tight for the Fe binding site. I think I will try just H295,H310,Q314 at that site to see if at least those 3 can be pulled together. And then try other residues for the fourth one-by-one. As for the Zn binding site, regarding Kevin's comments about the omission of H186 (and the apparent strong conservation of other polar residues in the 3D neighborhood), it looks to me like the N-terminus of the alignment (up to about residue 227 of T0204) is effectively based on a single sequence with several instances -- the SCOP d.13.1.2 targets noted above. The two d.13.1.1 family members (targets as grouped above) are homologous to the C-terminus and thus are not relevant to the Zn binding site. So arguments from conservation of residues at this position may be due to artifacts like the inclusion of the N-terminal domain of 1ems, which is totally unrelated. Anyway, H133 and H184 seem to have easily been constrained to go well with C63 and C66. H186 seems like it is too far away. Furthermore, in 1guq, the histidines making up the binding site are from separate strands so I was inclined to do the same thing in the target, hence H133 and H184. My main concern with the Zn binding site is still the break at C63, which may be due to the sidechains of S61,S62 interfering with H133. Maybe increasing the break weight some more will help. So for try4: break (30 ->40) Fe binding site with just H295,H310,Q314 Tue Jun 22 16:31:54 PDT 2004 Kevin Karplus I superimposed all the best models using the superimpose-best.under script, and looked at the putative binding sites. They are almost identical in the different models. Using the "try1" cost function (which has th eleast manual input), currently try2-opt2 scores the best. The hardest thing for us may be to get the linker S38-K88 properly packed against the main body of the protein. I think that clustering C63, C66, H133, H184 is probably a good idea. Clustering C216, C219, H255, H310 is probably also a good idea. Either C73 or H83 (or both) may be connecting H124 and H186. We may want to look for possible pairings that will put the linker down on the surface---if it is not a disordered region. For example, we could try to put F48 near F248 and P45 near G31 and P56 near L278. These should not be very strong constraints---they're just trying to pull the linker down to the surface. Wed Jun 23 12:11:48 PDT 2004 Sol Katzman The try4 result did repair the break at C63 near the Zn binding site. No obvious improvement at the Fe binding site. For try5, take Kevin's suggestion for the Fe binding site, and use C216, C219, H255, H310 there. For the linker, I examined the 1guq homolog and found that it has a close approach between a loop A307-T308-V309 and Q37-E38-T39. The analogous regions in T0204 are S320-G321-V322 and T46-D47-F48. I am going to try getting D47 near S320. Wed Jun 23 20:07:27 PDT 2004 Sol Katzman The results from try5 are encouraging. The linker pulled in at D47 as desired to the interior quite well. The main problems I see now are: -- the loop E104-G120 seems to stick out into space -- the backside of the 8-stranded sheet is predicted as burial but is largely exposed. That is, one side of this sheet is buried by the two helices P142-A160 and D264-K283, but the other side of this sheet is exposed. However, these two features are pretty much the same in 1guqA, which suggests that this is an active site. Note that in 1guqA, the Fe binding site covers part of this side of the sheet. Wed Jun 23 20:28:55 PDT 2004 Kevin Karplus Using the try5 cost function, try4-opt2 is the best model, slightly better than try5-opt2. Using an unconstrained cost function, try4-opt2 is still the best, but now try2 is second best, rather than try5. try5 looks good in a quick eyeball check, but has somewhat worse clashes than try4. Clashes and breaks are still quite large---perhaps it is time for a polishing run using the unconstrained cost function (after a Rosetta repack of try5-opt2). This should start with TryAllAlign commented out and ReadConformPDB for the try5-opt2 and try5-opt2.repack-nonPC. (including other models may cause the wrong one to be polished). Turning up the rotamer methods and crossover may be a good idea, to make clashes more likely to be resolved by repacking than by pulling the model apart. If the polishing run blows thing up, then another constrained run from the best so far may be in order. If it comes out better than try4, then a final polishing run from all the models (and repacked versions of the better ones), would be useful to do a final cleanup. Thu Jun 24 12:59:29 PDT 2004 Sol Katzman Polishing run try6 submitted with suggestions from Kevin: ReadConformPDB T0204.try5-opt2.pdb ReadConformPDB T0204.try5-opt2.repack-nonPC.pdb try6.costfcn <- unconstrained.costfcn for opt1: InitMethodProbs IS WAS OneRotamer 4 1 ClashingRotamer 4 1 ClusteredRotamer 5 2 CrossOver 6 3 for opt2: InitMethodProbs IS WAS OneRotamer 8 5 ClashingRotamer 3 1 ClusteredRotamer 3 1 CrossOver 3 1 Fri Jun 25 10:53:54 PDT 2004 Sol Katzman Try6 looks at least as good as try 5, and has a much better score than previous tries, using unconstrained.costfcn (probably due to the reduces clashes) but has many breaks. The best are: try6-opt2 (173 clashes 22 breaks) try6-opt1 (229 clashes 17 breaks) try4-opt2 (301 clashes 11 breaks) try2-opt2 (243 clashes 15 breaks) try1-opt2 (248 clashes 9 breaks) try5-opt2 (330 clashes 9 breaks) For try7 polishing run, include these models: opt2 try1,2,3,4,5,6 opt2 repack-nonPC try 4,5,6 opt1 try1,2,3,4,5,6 opt1 repack-nonPC try 6 Sat Jun 26 09:28:12 PDT 2004 Kevin Karplus try7-opt2 is new best scorer, with try6-opt2 not too far behind. Rosetta doesn't like any of the repacked models, but hates try6-opt2 least. For try7-opt2.repack-nonPC, the residues that Rosetta hates most are all seen by undertaker as clashes: other-bump: 1.88357 Ang (T0204)P245.CB and (T0204)L319.CD2 threshold= 3.07655 cost= 0.947338 other-bump: 2.07772 Ang (T0204)L319.CB and (T0204)G323.C threshold= 3.17773 cost= 0.921869 319 LEU -3.6 655.1 1.5 0.0 -0.3 3.2 0.5 0.0 0.0 -0.1 0.0 0.0 656.6 245 PRO -2.8 432.9 0.6 0.0 -0.3 0.2 0.0 0.0 0.0 0.0 0.0 0.0 430.6 323 GLY -2.0 223.8 1.1 0.0 1.1 -0.0 0.0 -0.2 0.0 0.6 0.0 0.0 223.2 other-bump: 2.25485 Ang (T0204)I248.N and (T0204)I315.CD1 threshold= 3.06506 cost= 0.841485 other-bump: 2.40335 Ang (T0204)Y292.CD1 and (T0204)I315.CA threshold= 3.20265 cost= 0.821415 315 ILE -5.4 182.8 1.6 0.0 -0.7 4.6 11.8 -1.0 0.0 -0.3 0.0 0.0 194.0 248 ILE -5.0 106.8 1.6 0.0 -0.1 2.1 0.1 -1.0 0.0 -0.3 0.0 0.0 104.8 292 TYR -7.8 120.6 3.0 0.0 -0.5 2.5 0.6 -1.0 0.0 -0.8 0.0 0.0 118.2 other-bump: 2.04154 Ang (T0204)N86.CB and (T0204)S130.CB threshold= 3.15526 cost= 0.926626 130 SER -3.7 141.5 2.7 0.0 -0.6 0.4 0.0 -0.1 -0.2 0.6 0.0 0.0 139.6 86 ASN -3.9 126.3 3.1 0.0 -0.8 2.8 2.0 0.0 -0.2 0.9 0.0 0.0 128.5 other-bump: 2.10343 Ang (T0204)L103.O and (T0204)F121.CE1 threshold= 2.95529 cost= 0.869994 other-bump: 2.24273 Ang (T0204)F121.O and (T0204)P193.CD threshold= 2.83689 cost= 0.755875 121 PHE -5.5 96.2 1.8 0.0 -0.3 0.4 0.1 0.0 0.0 -1.0 0.0 0.0 93.7 193 PRO -3.4 58.4 1.0 0.0 -0.1 0.1 0.0 0.0 0.0 0.0 0.0 0.0 56.0 neighbor-bump: 1.65296 Ang (T0204)D287.O and (T0204)P288.CD threshold= 2.49479 cost= 0.915401 288 PRO -2.2 54.8 0.9 0.0 2.5 2.1 0.0 0.0 0.0 0.0 0.0 0.0 58.0 other-bump: 1.80321 Ang (T0204)S257.O and (T0204)T296.OG1 threshold= 2.41651 cost= 0.827356 296 THR -3.5 49.8 2.8 0.0 -0.5 0.8 0.0 -0.7 -0.2 0.4 0.0 0.0 48.1 other-bump: 2.10878 Ang (T0204)D84.O and (T0204)P131.CG threshold= 2.68408 cost= 0.764814 131 PRO -4.0 44.5 1.3 0.0 -0.5 0.8 0.0 0.0 0.0 0.0 0.0 0.0 42.2 Perhaps another polishing run, with soft_clashes turned up to around 50? Some of the dry weights can be turned up to keep the alignment from being spread out too much. I'll try this as try8. Sat Jun 26 15:30:14 PDT 2004 Kevin Karplus try8-opt2 was optimized without constraints, and has the new best score with the try8.costfcn. Rosetta likes the repacked try8 best also, though it hates them all. The cluster of CYS at C216,C217,C219 are all pointing in different directions. This is probably wrong, but is it worth fussing with? Maybe we should turn the constraints back on for the Fe and Zn binding sites and try fixing rotamers by optimizing from the best of the current models? The loop from E104-G120 is floating around out in space, but that comes from the alignment. I should check this structure in the dimer from 1hxp. Ah--I see from dimeric-1hxp-try8.psdb that the loop is actually a hairpin that forms the last two strands of a sheet in the dimeric structure. Should I add strand constraints? Would a SheetConstraint be appropriate? From karplus@soe.ucsc.edu Sat Jun 26 23:02:54 2004 Date: Sat, 26 Jun 2004 23:02:50 -0700 From: Kevin Karplus To: katzman@ieee.org CC: sol@soe.ucsc.edu, ggshack@soe.ucsc.edu, learithe@soe.ucsc.edu, martina@soe.ucsc.edu, bbarnes@ucsc.edu, marcias@ucsc.edu, rph@soe.ucsc.edu, karplus@soe.ucsc.edu In-reply-to: <5.1.1.6.1.20040626221438.00b1ec18@pop.sbcglobal.yahoo.com> (message from Sol Katzman on Sat, 26 Jun 2004 22:28:09 -0700) Subject: Re: T0204 and T0205 Sol asked, > About T0204 and the loop that sticks out: > > Where does dimer-1hxp-try8.pdb come from? I also looked at 1guq.pdb, > using restrict (:a or :b). In the 1guq file, part of the loop is shown to > form a sheet with the other monomer, but in the dimer-1hxp file, the > loop seems to pass through the helix at the N-term of the other > monomer. The dimer comes from superimposing our monomer on both chains of the 1hxp dimer---see the Makefile and the make-dimer-try8.under script. (One could get a similar result by superimposing on the 1guq.pdb file.) The N-terminal helix of T0204 in try8 *is* badly placed, since it conflicts with the hairpin of the loop that should continue the sheet of the other monomer. We'll need to pin it back by adding some constraints---it probably packs next to the C-terminal helix. Maybe make A7 be somewhere near V344?? ------------------------------------------------------------ Mon Jun 28 11:57:29 PDT 2004 Sol Katzman In the 1guq dimer, there is only a very short part R98-S101 of the loop that forms a strand bonding to the sheet in the other monomer. The conformation of the rest of the loop does not look like it is part of the sheet. So let's just put a strand constraint for T0204 R116-V119 To get the N-terminal helix out of the way of the other monomer, take Kevin's suggestion to constrain A7 to V344. Also, turn on the Zn and Fe binding constraints from try5. Zn C63, C66, H133, H184 Fe C216, C219, H255, H310 Include these models opt2 try1,2,3,4,5,6,7,8 opt2 repack-nonPC try 4,5,6,7,8 opt1 try1,2,3,4,5,6 opt1 repack-nonPC try 6 This will be try9. ------------------------------------------------------------ Tue Jun 29 13:01:52 PDT 2004 Sol Katzman There was a power outage that killed try9 so we only have opt1 and opt1-scwrl results. We did not get a strand on the intended loop, but it does seem well placed in the dimer-1hxp-try9.opt1-scwrl.pdb Also, the N-terminal helix now avoids the collision in the dimer. The Zn binding site looks okay but the Fe binding site is not very tight. I am not convinced we have the proper residues for the Fe site, but they are what we have been using since try3. The best scores using the try9 constraints are: clashes breaks T0204.try9-opt1.pdb 128 22 T0204.try9-opt1.repack-nonPC.pdb 120 22 T0204.try8-opt2.pdb 124 23 T0204.try9-opt1-scwrl.repack-nonPC.pdb 163 22 T0204.try7-opt2.pdb 160 22 T0204.try8-opt1.pdb 153 22 T0204.try7-opt1.pdb 171 22 T0204.try6-opt2.pdb 173 22 T0204.try8-opt2.repack-nonPC.pdb 124 23 T0204.try6-opt2.repack-nonPC.pdb 133 22 Tue Jun 29 13:48:24 PDT 2004 Kevin Karplus try9-opt1 scores best currently. Rosetta likes try8-opt2.repack-nonPC best. I'm not real fond of how I67-W87 is floating of by itself in try9-opt1. The N-terminal helix is tucked out of the way, but not packed well against the protein. Maybe R10 should saltbridge with E347??? Now that the tweak operators are working properly, it may be worth doing another polishing run, perhaps with an extra constraint on the N-terminal helix. Tue Jun 29 16:36:52 PDT 2004 Sol Katzman Concerning the loop I67-W87, and in particular whether the second half of this should get closer to the strand that starts where this ends, I looked at the 1guqA template. In 1guqA, there is no Hbonding along V61-T76 (which corresponds to roughly E76-L95 in T0204), and looking at these residues in spacefill mode there is still daylight between them. So I don't want to try to pack this in any more. I did add a saltbridge constraint for the N-terminal and C-terminal helices, slightly different than the one Kevin suggested: Hbond D9.OD1 R346.NH2 1.0 Hbond D9.OD2 R346.NH1 1.0 This will be try10. Fri Jul 2 12:50:01 PDT 2004 Sol Katzman The results of try10 just seem to have broken the N-terminal helix. It did not get any closer to the C-terminal helix. Fri Jul 2 15:23:29 PDT 2004 Martina Koeva I changed the salt bridge constraint for the N-terminal and C-terminal helices to: Hbond R10.NH1 E339.OD2 1.0 Hbond R10.NH2 E339.OD1 1.0 so that the "front end" of the C-terminal helix can pair in an anti-parallel way with the N-terminal helix. Some additional observations: In top templates: The first 10 residues that seem to form a helix in T0204 are an insertion that we do not find in those templates: 1hxqA/B, 1hxpA, 1guqA. The only templates that seem to align in that segment are the human, mouse, rat galactose-1-phosphate uridylyltransferase, which we do not have structures for and none of the literature seems to give any information about this initial N-terminal insertion. Sat Jul 3 08:14:28 PDT 2004 Kevin Karplus The unconstrained costfcn scores try11-opt2 best so far, with try8-opt2 and try10-opt2 not too far behind. I'll try repacking try11 and try10, to see what Rosetta thinks of them---it likes try10-opt2.repack-nonPC best. I don't want to include try8 in the "best models" because of the bad clash in the dimer. With the try12 cost function, try12-opt1 scores best (try12 not finished yet), and try11-opt2 close behind, and try10-opt2 behind that. I like the way that try12-opt1 is packing the N-terminal helix, and I hope that try12-opt2 improves it further, so I can move it up in the ranking. With the try11 cost function, the order is try11-opt2, try10-opt2, try9-opt1, try12-opt1. Sat Jul 3 08:47:50 PDT 2004 Kevin Karplus try12 just finished--I'll repack it and see what Rosetta thinks of it also. Ah--it likes it better than try10-opt2 repacked, though it thinks that all the models have bad clashes. Undertaker still likes try11-opt2 better than try12-opt2, but both are doing better than try8-opt2. I'll try an unconstrained optimization run, starting from all the current models. Sat Jul 3 20:08:00 PDT 2004 Kevin Karplus I repacked and scored try13. Rosetta does not like the repacked version as well as it likes try12-opt2.repack-nonPC. Unfortunately, try13 has a bad clash between the helix and the hairpin from the dimeric partner, so it must be rejected. I think we'll submit 1 T0204.try11-opt2.pdb 2 T0204.try12-opt2.pdb 3 T0204.try12-opt2.repack-nonPC.pdb 4 T0204.try10-opt2.pdb 5 T0204.t2k.undertaker-align.pdb model 1 T0204-1hxpA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m Sun Sep 19 12:09:48 PDT 2004 Kevin Karplus I added REAL_PDB:=1vkvA FINAL_COSTFCN:=try13 to the Makefile and did a whole-chain rmsd evaluation of our models. Best rmsd for model5, robetta5, robetta1, model2, model3, model1, model4, robetta3, robetta2, robetta4. Model5 has artificially good rmsd because of being incomplete. Foo, it looks like robetta beat us, though the differences are small. Rosetta repacking did not make model3 better than model2! Wed Sep 22 10:50:20 PDT 2004 Kevin Karplus With GDT score the order is try4-opt2 69.4719% model5 68.2343% model1 67.9043% model3 67.8218 robetta3 67.8218 model2 67.5743 Something wrong here: model2 and 3 have same backbone. model4 67.2442 robetta1 66.5842 robetta4 66.5017 robetta5 65.5116 robetta2 64.6040 We beat robetta, but our best model was just sidechain replacement on the template. Wed Sep 22 19:55:06 PDT 2004 Kevin Karplus The problem above with GDT is inherent to the GDT definition. By using "smooth_GDT", which I set to be a bit more stringent, I get much smaller differences: model5 53.6894 model1 52.6375 model2 52.4514 model3 52.4505 model4 52.0588 robetta3 51.7963 robetta4 51.0846 robetta5 50.7935 robetta1 50.7636 robetta2 49.7578 Fri Sep 24 16:55:44 PDT 2004 Kevin Karplus The GDT curves for all the robetta and submitted models are very similar, but the model5 curve gets a few more points in the 2-6 Ang region, crossing over and getting fewer above 8Ang. The new parameters for smooth_GDT make try4-opt1 the best name length missing_atoms rmsd rmsd_ca GDT smooth_GDT T0204.try4-opt1.pdb.gz 351 0.0000 4.3416 3.9467 -67.8218 -64.0006 model5.ts-submitted 351 274 3.5851 3.1259 -66.7492 -62.9815 model1.ts-submitted 351 0.0000 4.7767 4.3831 -66.5842 -61.9036 model2.ts-submitted 351 0.0000 4.7748 4.3370 -65.5116 -60.9810 model3.ts-submitted 351 0.0000 4.7764 4.3370 -65.4290 -60.9727 model4.ts-submitted 351 0.0000 4.8617 4.5019 -64.1914 -59.1316 robetta-model5.pdb.gz 351 0.0000 4.5461 4.0482 -63.3663 -58.5962 robetta-model3.pdb.gz 351 0.0000 5.1480 4.7901 -63.1188 -58.4817 robetta-model1.pdb.gz 351 0.0000 4.6265 4.1231 -62.9538 -58.4252 robetta-model4.pdb.gz 351 0.0000 5.5536 5.4052 -61.9637 -58.1289 robetta-model2.pdb.gz 351 0.0000 5.2792 4.7648 -60.9736 -56.7660