Mon Jul 17 09:35:43 PDT 2006 T0381 Make started Mon Jul 17 09:36:18 PDT 2006 Running on lopez.cse.ucsc.edu Mon Jul 17 09:49:05 PDT 2006 Kevin Karplus BLAST gets full-length hit on 2g7uA (43% gapless over 246 residues, e-value 2e-43) This looks like a close homology modeling case---everyone will get the template and alignment, and we won't do better than average on refinement (maybe worse as we don't use a frozen core). Mon Jul 17 16:10:27 PDT 2006 Kevin Karplus Interestingly, the HMMs prefer 1mkmA somewhat to 2g7uA. This may just be drift in the multiple alignments, since BLAST sees 1mkmA as quite a bit more distant (28% id oevr 155 residues, E-value 5.5e-14). Mon Jul 24 14:53:14 PDT 2006 Crissan Harris, Cynthia Hsu After examining the best-models, it seems pretty clear that residues M1-S91 do not coincide at all with the other alignments. In view of this, we've decided that it may be to our advantage to produce a subdomain and then form a chimera superimposing it in a similar fashion. The subdomain is currently being made on lopez. Tue Jul 25 11:34:06 PDT 2006 Crissan Harris, Cynthia Hsu The best models for our subdomain seem to generate alignments that match our try1 very well, with the exception of residues M1-R22. The new subdomain also appears to be an improvement over the subdomain produced by the alignment of the entire target, as it is more compact, with shorter helices. After chimera1-1.pdb was formed from the 91 atoms of the subdomain and the remaining atoms of try1-opt2, we made the try2.under file to ReadConformPDB chimera1-1.pdb. In try2.costfcn, we commented out the dssp-ehl2 constraints, the undertaker alignments, and the rr constraints, and set "soft_clashes" to 30 and "break" to 80. try2 is currently running on shaw. Tue Jul 25 15:02:35 PDT 2006 Crissan Harris, Cynthia Hsu try2 produced a very nice-looking chimera that scores very well. We were curious, however, as to why none of the residues from M1-R79 from our two tries were aligning with undertaker's top models, so we decided to attempt to produce a third and fourth try by restricting alignment to the first and second models. For try3, we had it read in only the first undertaker alignment, 1mkmA. We used the try1.costfcn. However, trying to run try3 failed. We decided to copy everything to try4 and repeat the process. try4 is currently running on camano. try5 used the second alignment, 2g7uA, as its only alignment. We also used a copy of try1.costfcn for try5.costfcn. try5 is currently running on shaw. Wed Jul 26 09:11:08 PDT 2006 Crissan Harris try4 and try5 both look pretty decent. try4 has one significant break and try5 has about three. Both of the models look okay under the near script but under the burial script they have some regions that are predicted to be buried but are exposed. I definitely like the overall shape of try4 better than try5, especially since try5 is very foamy and has one significantly large hole. Also in superimposing them with the best models it is clear that neither try4 nor try5 align the helices before R79 to the helices in the alignment models. I looked at the sequence alignment for our sequence and those of 1mkmA and 2g7uA and it is clear that they do not have much alignment in these begining residues. None the less, try4 and try5 are still different from tries 1 and 2 so it gives us some variation. Since try4 and try5 do not score as well as try1 and try2 I will try to run another try on each of these alignments trying to improve the breaks and the packing. try6 uses 1mkmA for the alignment. the dry weights were modified as follows: dry5 20 dry6.5 25 dry8 20 dry12 6 "n_ca_c" was raised to 7, "soft_clashes" was raised to 40, and "break" was raised to 80. try6 is currently running on orcas. try7 uses 2g7uA for the alignment. the dry weights were modified as follows: dry5 20 dry6.5 25 dry8 20 dry12 6 "phobic_fit" was raised to 4, "n_c_ac" was raised to 6, "bad_peptide" was raised to 11, "soft_clashes" to 50 and "break" to 100 try7 is currently running on shaw. Wed Jul 26 14:33:48 PDT 2006 Cynthia Hsu Apparently tries 6 and 7 did significantly worse than their predecessors. This may be because the .under files were set up to generate new models from the same alignments but with higher weights on unimportant consstraints. I've decided to attempt polishing runs of the existing models (try4 and try5) but using the cost functions that Crissan did. In particular, there was only one significant break for try4, which was only slightly above a cost of one. In view of this, I copied try6.costfcn to try8.costfcn and commented out the dssp-ehl2 constraints, the undertaker alignments, and the rr constraints. try8.under used try4-opt2 for its ReadConformPDB. try8 is currently running on squawk. I was not sure whether it was better to do a polishing run of try7 or try5. Both looked extremely foamy, but our attempts to change it just by modifying Undertaker constraints seem to generate worse-scoring models. I agree wtih Crissan that try5 has an extremely large hole that should probably be taken care of. However, try7 has several smaller holes, so I am not entirely sure which one is better for polishing. I've decided on try7, but I am not entirely sure if this is a good idea. Wed Jul 26 15:20:18 PDT 2006 Cynthia Hsu For try9.costfcn, I used try7.costfcn. I raised "dry12" to 8, however, hoping that this would decrease the foaminess. "soft_clashes" was raised to 60 and "break" is raised to 150. I also commented out the dssp-ehl2 constraints, the undertaker-alignments, and the rr constraints. try9 is currently running on squawk. Wed Jul 26 21:40:06 PDT 2006 Cynthia Hsu My polishing runs seemed to have made the models significantly worse, instead of better. I think I will refrain from doing any more work on this at the moment. try2 is currently the most favored model by Rosetta. Thu Jul 27 11:00:46 PDT 2006 Crissan Harris, Cynthia Hsu On closer examination, we found that try8.under and try9.under took in T0379.try4-opt2 and T0379.try7-opt2 as their input PDB files. This was changed. try10, which took T0381.try4-opt2.pdb as the input file, is currently running on shaw. The costfcn was identical to that of try8. try11, which took T0381.try7-opt2.pdb as the input file, is currently running on lopez. The costfcn was identical to that of try9. Thu Jul 27 13:58:06 PDT 2006 Crissan Harris, Cynthia Hsu Both try10 and try11 did fairly well in terms of unconstrained scores, although still not as well as tries 1 and 2. try11, although the last favored of the four by Undertaker, was the most favored by Rosetta. In view of this, we decided to do a polishing run of the gromacs repacked version of try11. We copied try11.costfcn to try12.costfcn, then raised "dry5" to 25, "dry6.5" to 35, and "dry8" to 25. "n_ca_c" was raised to 10, and "bad_peptide" to 15. "soft_clashes" was raised to 80 and "break" to 200. try12 was set up as a polishing run of the repacked gromacs version of try11. try12 is running on camano. We also decided to do a polishing run on try1. We copied the try1.costfcn, then commented out the dssp-ehl2 constraints, undertaker alignments, and rr constraints. We raised "dry5" to 20, "dry6.5" to 25, and "dry8" to 20. We also raied "n_ca_c" to 7, and "bad_peptide" to 12. "soft_clashes" was raised to 60 and "break" was raised to 150. try13.under used T0381.try1-opt2 for the ReadConformPDB. try13 is currently running on cheep. Thu Jul 27 16:32:40 PDT 2006 Crissan Harris, Cynthia Hsu try12, surprisingly, performed the worst out of all present models. try13, however, did successfully polish try1-opt2. We've decided that we have generated as many diverse models as possible for this target, and thus have decided to do a generic polishing run, using "include read-pdb.under". Since try2 was our top scoring model, we copied try2.costfcn to try14.costfcn, then rasied "dry5" to 25, "dry6.5" to 30, and "dry8" to 20. "soft_clashes" was raised to 60, "break" was raised to 150, "hbond_geom" raised to 6, "hbond_geom_backbone" raised to 12, "hbond_geom_beta" raised to 60, "hbond_geom_beta_pair" raised to 110. We also turned off "maybe_ssbond" due to the absence of cys residues in close proximity to each other. In addition, "n_ca_c" was raised to 7 and "bad_peptide" to 12. Fri Jul 28 10:33:01 PDT 2006 Crissan Harris, Cynthia Hsu try14 was our top scoring model. However, we are still having many problems with breaks and soft clashes in all our models. try14 (gromacs repacked) was also the model most favored by Rosetta. We copied try14.costfcn to try15.costfcn, then raised "dry5" to 30, "dry6.5" to 35, "dry8" to 20. We lowered "sidechain" to 4, then raised "n_ca_c" to 10, and "bad_peptide" to 12. try15.under was set to ReadConformPDB T0381.try14-opt2.gromacs0.repack-nonPC.pdb.gz try15 is currently running on shaw. Fri Jul 28 12:18:32 PDT 2006 Crissan Harris, Cynthia Hsu try15 is now the top favored model by Rosetta, and scores fairly well in Undertaker. We have decided that our top five models are as follows: ReadConformPDB T0381.try14-opt2.pdb //Most favored by Undertaker - chimera, from try2 #ReadConformPDB T0381.try2-opt2.pdb // top alignment before try14; identical ReadConformPDB T0381.try15-opt2.gromacs0.repack-nonPC.pdb // very similar to try14, but most favored by Rosetta ReadConformPDB T0381.try13-opt2.pdb //initial alignment, polished. ReadConformPDB T0381.try10-opt2.pdb //used 1mkmA as the alignment (polished) ReadConformPDB T0381.try11-opt2.gromacs0.repack-nonPC.pdb //used 2g7uA as the alignment (polished) Sun Jul 30 15:28:24 PDT 2006 Cynthia Hsu Despite its impressive score, the soft_clashes, breaks, n_ca_c, and bad_peptide scores for try14-opt2 can still be greatly improved. In view of this, I raised "n_ca_c" to 10, "bad_peptide" to 15, "soft_clashes" to 80, and "break" to 180. try16 uses ReadConformPDB T0381.try14-opt2.pdb. try16 is currently running on shaw. Mon Jul 31 11:17:10 PDT 2006 Cynthia Hsu There was no improvement in the overall cost of "break". For try17.costfcn, I decided to raise the weight of "soft_clashes" to 100 and that of "break" to 210. I also decided to use "include read-pdb.under". try17 is currently running on vashon. Mon Jul 31 15:41:12 PDT 2006 Crissan Harris, Cynthia Hsu try17 was much improved in breaks and soft_clashes, but scored poorly in other areas. try17.costfcn was copied to try18.costfcn. "dry5" was raised to 28, "dry6.5" to 35, and "bad_peptide" to 20. "hbond_geom" was lowered to 5, "hbond_geom_backbone" to 10, "hbond_geom_beta" to 50, and "hbond_geom_beta_pair" to 100. try18.under used try17-opt2 as its input file. try18 is currently running on lopez. try13 also did badly on breaks, so its cost function was copied to try19.costfcn. "break" was raised to 200, and "soft_clashes" to 80. "dry5" was raised to 25, "dry6.5" to 30, "and dry8" to 25. "n_ca_c" was raised to 10 and "bad_peptide" to 15. "maybe_ssbond" was removed from the cost function. try19.under used ReadConformPDB T0381.try13-opt2.pdb try19 is currently running on camano. Tue Aug 1 10:22:55 PDT 2006 Crissan Harris try18 is now the top scoring model in Undertaker's unconstrained scores as well as try18-opt2.gromacs0.repack-nonPC.pdb.gz being Rosetta's favorite. try19 scores worse than many of our models but did manage to do better than try13 (which it was a polishing run of). try18-opt2 does very well on breaks with all remaining braeaks having a cost less than 1. try19 still has some large breaks. T0381.try19-opt2.gromacs0.repack-nonPC.pdb.gz has the least amount of breaks as well as the lowest costs. With this in mind I have decided to do another polishing run of try19 reading in the T0381.try19-opt2.gromacs0.repack-nonPC.pdb.gz version in the hopes of reducing the breaks. try20.costfcn was copied from try19.costfcn with the following changes: dry5 30 dry6.5 33 dry8 28 dry12 6 "sidechain" was lowered to 4, "n_ca_c" was raised to 13, "soft_clashes" was raised to 90, and "break" was raised to 235. try20 is currently running on vashon. Our new top models in best-models.pdb are as follows: ReadConformPDB T0381.try18-opt2.pdb //polished from try17 < polished from read-pdb.under; most favored by Undertaker and Rosetta #ReadConformPDB T0381.try15-opt2.gromacs0.repack-nonPC.pdb //3rd favorite of Rosetta, identical to try18 ReadConformPDB T0381.try14-opt2.pdb //4th favorite of Undertaker, 4th favorite of Rosetta, very similar to try18 ReadConformPDB T0381.try19-opt2.pdb //try13 polished < initial alignment polished. ReadConformPDB T0381.try10-opt2.pdb //used 1mkmA as the alignment (polished) ReadConformPDB T0381.try11-opt2.gromacs0.repack-nonPC.pdb //used 2g7uA as the alignment (polished) Tue Aug 1 13:19:39 PDT 2006 Crissan Harris try20 scored worse than try19 but did manage to close a lot of the breaks, however one large break still remains. I have decided to do a polishing run of try20-opt2.gromacs0.repack-nonPC.pdb.gz to attempt to close these final breaks. The try21.costfcn was copied from the try20.costfcn and the following changes were made: dry5 33 dry6.5 37 dry8 32 dry12 6 break 275 try21 is currently running on vashon. Tue Aug 1 15:32:59 PDT 2006 Crissan Harris try21 still scored worse than try20 (as well as try19). However, it did manage to close the breaks. There are now only a few breaks left that all have a cost under 0.65. Thus I will try one more attempt to polish this model now that the breaks have closed. thus try22.costfcn was copied from try21.costfcn with the following changes: dry5 37 dry6.5 42 dry8 35 phobic_fit 3, n_ca_c 15, bad_peptide 17, soft_clashes 95, break 285 try22.under was set up to ReadConformPDB T0381.try21-opt2.pdb try22 is currently running on peep. Wed Aug 2 10:31:11 PDT 2006 Cynthia Hsu try22 is still pretty low scoring. I've raised "dry5" to 40 and "dry8" to 37. "n_ca_c" was raised to 17 and "bad_peptide" to 20. I raised "sidechain" to 5. I also raised "constraints" to 12. try23.under was set up to ReadConformPDB T0381.try22-opt2.pdb try23 is currently running on vashon. Wed Aug 2 14:31:30 PDT 2006 Crissan Harris, Cynthia Hsu Our current top models are as follows: Model 1 is try18-opt2.pdb. This was polished from the following: try18-opt2 < try17-opt2 < try15-opt2 < try14-opt2.gromacs0.repack-nonPC < try2-opt2 < chimera1-1 chimera1-1 was made from the try1-opt2 model generated by Undertaker alignments (1mkmA, 1tf1A, 1ysqA, 2cfxA, 1r1tA, 1j75A, and 1r7jA) for residues 92-246. Residues M1-S91 were superimposed from a separately created subdomain (from alignments to 2g7uA, 1mkmA, 1r7jA, 1smtA, 2cg4A, and 2fnaA). This model was the most favored by both Undertaker and Rosetta, and polished a model that had previously been one of Rosetta's favorites (the gromacs repacked version of try14). Model 2 is try16-opt2.pdb. This was polished from the following: try16-opt2 < try14-opt2 < try2-opt2 < chimera1-1 chimera1-1 was made from the try1-opt2 model generated by Undertaker alignments (1mkmA, 1tf1A, 1ysqA, 2cfxA, 1r1tA, 1j75A, and 1r7jA) for residues 92-246. Residues M1-S91 were superimposed from a separately created subdomain (from alignments to 2g7uA, 1mkmA, 1r7jA, 1smtA, 2cg4A, and 2fnaA). This model, although very similar to Model 1, is the second most favored by Undertaker Model 3 is try23-opt2.pdb. This was polished from the following: try23-opt2 < try22-opt2 < try21-opt2 < try20-opt2 < try19-opt2 < try13-opt2 < try1-opt2 < 1mkmA, 1tf1A, 1ysqA, 2cfxA, 1r1tA, 1j75A, and 1r7j We selected it because the intial alignment produced from Undertaker generated different models than would have been produced by our chimera. Model 4 is try10-opt2.pdb. This was polished from the following: try10-opt2 < try4-opt2 < 1mkmA We selected 1mkmA for several reasons: it produced a diverse fold, and the HMMs preferred it to 2g7uA (the top BLAST match). It also scored better than the polished model that had been generated from 2g7uA. Model 5 is try11-opt2.pdb. It was polished from the following: try11-opt2 < try7-opt2 < 2g7uA We selected 2g7uA for the alignment because it was the top match according to BLAST, with 43% sequence identity and e-value 2e-43. It produced a slightly different fold from the other five models, particularly across residues M1-S91. Thu Aug 3 07:16:26 PDT 2006 Kevin Karplus I think that we should have at least one model that follows the alignments without moving M1-S91. This region is still getting superb hits to 2g7uA, so I don't think there is a clear justification for packing it differently than the templates do. It looks like m1-S91 is a dimerization interface, so the modles should ahve been optimized as dimers. Thu Aug 3 07:55:36 PDT 2006 Kevin Karplus I set up a dimerization run. dimer/mult1 running on cheep. It will attempt to use double-length alignments to 1mkm and 2g7u to get the dimer interface right, but monmer alignments to do the finer work. Thu Aug 3 10:46:34 PDT 2006 Kevin Karplus Well dimer/try1 gets *a* dimer, but ti doesn't look like our top templates, so I'm going to try again, with even more focused dimer creation, using *only* 1mkm and 2g7u templates. Thu Aug 3 10:59:27 PDT 2006 Kevin Karplus dimer/mult2 started on cheep. Thu Aug 3 11:48:05 PDT 2006 Kevin Karplus dimer/mult3 started on lopez. Similar to try2, but with slightly different cost fcn and with some alignments other than 1mkm and 2g7u tossed in for the last 2 rounds of TryaAllAlign. If these don't work, we may have to get some scaffolding from the templates or from undertaker-align models. The alignment models are fine for constraints within a monomer, but we'd need to look at the dimeric templates to figure out what contraints to add between the monomers. Thu Aug 3 14:31:21 PDT 2006 Kevin Karplus RATS! both dimer/try2 and dimer/try3 blew the dimer apart. We're going to need intermonomer constraints. I'll make an incomplete dimer from 2g7uA+B and model 2 of T0381.undertaker-align.pdb.gz and another from 1mkmA+B and model 1 of T0381.undertaker-align.pdb.gz These incomplete dimers will serve as models to measure for creating constraints between the monomers. Thu Aug 3 15:05:41 PDT 2006 Kevin Karplus Problems: 1) I'm getting bad clashes in the dimer-1mkmA-1mkm and dimer-2g7uA-2g7u, which I did not expect. Perhaps undertaker isn't ignoring missing atoms? 2) The "dimers" seem to be tetramers---they need to be optimized in a tetramer context!! 2g7u is already a tetramer, but I'll have to pick up 1mkm.mmol from PQS. http://pqs.ebi.ac.uk/pqs-doc/macmol/1mkm.mmol Thu Aug 3 15:30:38 PDT 2006 Kevin Karplus To address the second problem, I set up a 4mer run starting from alignments This might still fail, so I should probably set up some incomplete 4mers to measure constraints from. I may be too tired to do this, though, so it might be worth someone else's time. Thu Aug 3 16:44:26 PDT 2006 Kevin Karplus I shouldn't be allowed to work when I'm this tired. I forgot to change "multimer 2" to "multimer 4" in 4mer/try1.under I killed 4mer/try1 and am trying again as 4mer/try2 on cheep. I still don't understand the clashes in the dimers from the undertaker-align models, unless chain A and chain B of the templates are not in the same conformation? THat is possible, but very problematic for us. Thu Aug 3 20:27:36 PDT 2006 Kevin Karplus RATS! I thought that 4mer/try2 had been running while I napped, but it had stopped on iteration 0, because the 4mer/T0381.a2m file still had only 2 copies of the chain, not 4. Thu Aug 3 20:38:35 PDT 2006 Kevin Karplus Trying again as 4mer/mult3 on cheep. I also had try3 output the try3-al4 model so that we can check earlier whether or not we are going to be optimizng a reasonable 4mer. Thu Aug 3 21:26:19 PDT 2006 Kevin Karplus Well, 4mer/decoys/T0381.try3-al4 looks quite reasonable, but it is clearly a dimer of dimers, not a cyclic 4mer. So I need to go back to the dimer scripts and fix them. Either that, or optimize the 4mer without using all the symmetries. Thu Aug 3 22:14:55 PDT 2006 Kevin Karplus I killed 4mer/try3 since the cyclizing had clearly ruined the oligomer in try3-opt1 I started dimer/mult4, to try to get the dimer right. I added a bund of constraints for the closest part of the dimer interface from 4mer/try3-al4 Unfortunately, I can't optimize that model directly, as it has a duplicated residue position, so gets atoms marked as missing. Fri Aug 4 01:33:51 PDT 2006 Kevin Karplus dimer/try4-opt2 begins to keep the right shape, but the D15-H32 helix needs some constraints also. Here are some distances (from trh 4mer try3-al4 model) L20.CB T59.CB(other) 4.3 Y16.CB R55.CB(other) 4.1 A21.CB T59.CB(other) 5.9 V26.CB F56.CB(same) 5.2 R22.CB T46.CB(same 5.8 V26.CB V42.CB 5.7 Fri Aug 4 02:25:04 PDT 2006 Kevin Karplus I've added constraints and started dimer/mult5 on cheep. I've also modified best-models.pdb.gz to include the best models so far. Fri Aug 4 04:16:34 PDT 2006 Kevin Karplus OK, dimer/try5 is beginning to look like the right model. There is a bit of separation between the two domains, but the dimer interface in domain1 looks pretty good. I think that the relation ship between the two domains is wrong---it is *not* the same as in the templates. I've started dimer/try6, which should concentrate more on 2g7u, rather than 1mkm. The superimpose-best.under file has been updated again. The buggy model from from-4mer-try3-al4 is still the closest to the templates. For dimer-try7, I should try adding some constraints involving the second domain (after V95). Here are some inter-monomer distances T4.CB T254.CB 6.7 P6.CB Q258.CB 7.0 E8.CB D255.CB 7.5 K9.CB R265.CB 6.5 A67.CB D259.CB 6.8 P77.CB A261.CB 3.6 E81.CB A219.CB 4.8 Y84.CB S220.CB 6.7 P98.CB D161.CB 9.2 Fri Aug 4 04:54:31 PDT 2006 Kevin Karplus dimer/mult7 started on lopez Fri Aug 4 06:30:26 PDT 2006 Kevin Karplus dimer/try6 scores better than dimer/try5 with the try5=try6 costfcn, but rosetta like try5 better. Fri Aug 4 07:13:00 PDT 2006 Kevin Karplus try7 scores well with the try7 costfcn, but still doesn't have the same structure as the templates. The N-terminal domain has flipped over! Fri Aug 4 07:32:27 PDT 2006 Kevin Karplus Let's try adding some scaffolding within a monomer to hold the first domain in place R28.CB A219.CB 11.9 D31.CB S219.CB 12.9 H32.CB S220.CB 11.0 D31.CB L260.CB 18.1 R44.CB T262.CB 27.0 D41.CB R217.CB 29.0 F30.CB E162.CB 24.7 Fri Aug 4 07:45:06 PDT 2006 Kevin Karplus dimer/mult8 (similar to mult7, but with scaffolding added to costfcn) started on cheep). Fri Aug 4 08:16:57 PDT 2006 Kevin Karplus dimer/mult9 started on lopez, with same constraints as try8, but with the scaffolding and inter-monomer constraints given a much higher weight. Fri Aug 4 12:03:17 PDT 2006 Kevin Karplus dimer/try8 and dimer/try9 come closest to the template dimers, but still have some rotation of the first domain relative to the second domain. Fri Aug 4 12:50:25 PDT 2006 Kevin Karplus I don't think I have the time or the energy to make a chimera of good structures with template-like geometry, so I'll just polish the existing dimers (which will probably concentrate on dimer/try8-opt2 to improve clashes and breaks. The dimers we are building are probably not close enough to the templates to make good dimer-of-dimer tetramers. We'll probably get the individual domains ok, but not the relationship between them. dimer/mult10 running on cheep. Fri Aug 4 19:39:30 PDT 2006 Kevin Karplus try10-opt2 does not look too bad. I'm going to submit InFilePrefix dimer/decoys/ ReadConformPDB T0381.try10-opt2.unpack.pdb chain A ReadConformPDB T0381.try7-opt2.unpack.pdb chain A ReadConformPDB from-4mer-try3-al4.pdb chain A # duplicate position, so incomplete InFilePrefix decoys/ ReadConformPDB T0381.try18-opt2.pdb InFilePrefix ReadConformPDB T0381.undertaker-align.pdb model 2 #2g7uA t06 Fri Aug 4 20:01:20 PDT 2006 Kevin Karplus Submitted with comment T0381 got two excellent fold-recognition hits (1mkmA and 2g7uA). The optimization was initially done under the supervision of two undergrads, who did not notice that these two templates were dimer-of-dimer tetramers, and so allowed undertaker to fold the two domains into a single compact monomer. Just days before the target expired, the mistake was noticed, and the protein was optimized in a dimer context. This turned out to be rather difficult, as we did not have a good monomer to superimpose on a dimer to begin the optimization. Instead we had to start from double-length alignments of both monomers simultaneously---a rather awkward way to create a dimer. Infact, undertaker still wanted to modify the relationship between the domains to get more contact dimer. We had to add a number of inter-monomer constraints and "scaffolding" constraints to keep the domains in roughly the orientation in the two templates. Even with the constriants, the N-terminal domain still wanted to rotate a bit into a different relationship with C-terminal domain. Since the two templates have slightly different relationships between the domains, the we decided that this was plausible. There was not time to optimize further in a proper tetramer context. Model 1 is chain A of dimer try10-opt2, optimized by undertaker from dimer/try8-opt2.unpack.gromacs0.repack-nonPC, which was dimer/try8-opt2 reoptimized by gromacs with sidechains repacked by rosetta. It was the best-scoring dimer with an unconstrained costfcn. dimer/try8-opt2 was optimized from alignments. Model 2 is chain A of dimer/try7-opt2, optimized from alignments by undertaker, but with fewer constraints than for try8-opt2. Model 3 is chain A of from-4mer-try3-al4, a model from alignments to 1mkmA, with all residues present, but some duplication of atom locations. Because of having multiple atoms in the same location, it is not optimizable directly by undertaker. Model 4 is monomer try18-opt2, the best-scoring model produced by the undergrads. It is included on the aoff chance that T0381 really does form a monomer rather than a dimer or tetramer. Model 5 is sidechain replacement by SCWRL on an alignment to 2g7uA. ----------------------------------------------------------------------