Wed Jun 11 08:25:49 PDT 2008 T0443 Make started Wed Jun 11 08:26:23 PDT 2008 Running on cheep.cse.ucsc.edu Sat Jun 14 13:49:40 PDT 2008 Kevin Karplus Somehow, whatever notes I had on T0443 got wiped out. I think I ran a "new-target" by mistake. I'll have to fix that script to be less dangerous. I don't even have a backup copy on my machine at home, so I have to look at the results all over again. First, there are no good hits (best evalue is 2e1nA at E-value 21.5). Second, the try1-opt3 run is not particularly compact. The MQAC quality assessment prefers SAM-T08-server_TS1 and Zhang-Server_TS3 The MQAU quality assessment prefers three of the Zhang-Server models than SAM-T08-server_TS1. The predicted GDT is low (27%-36%). The conservation signals are very strong, though slightly different in the t2k and t06 alignments, so if there were a member of this family in PDB, the HMMs would have found it. (There could, of course, be a member of the fold in PDB that does *not* have the conserved residues, and so would be hard to find.) The MQAU1 and MQAC1 metaserver runs are mainly based on SAM-T08-server_TS1, so they might not provide much insight into other possibilities. I might want to look directly at the Zhang-Server models. Wed Jun 18 17:02:13 PDT 2008 Firas Khatib We are going to check the servers models and extract segments that we feel are decent. Just as an example of this, Kevin and I agreed that residues 73-117 in Zhang_Server_TS3 pack pretty well against each other. Chirag, John and I will investigate the other top-scoring server models according to SAM-T08-MQAC.qa1 and bring our results on friday at 3pm. Thu Jun 19 12:00:20 PDT 2008 SAM-T08-MQAO hand QA T0443 Submitted Thu Jun 19 12:00:20 PDT 2008 SAM-T08-MQAU hand QA T0443 Submitted Thu Jun 19 12:00:20 PDT 2008 SAM-T08-MQAC hand QA T0443 Submitted Thu Jun 19 14:26:21 PDT 2008 Firas Khatib looking at PSI_TS1, I quite like residues 174-222 because the 3 helices pack well against each other and the secondary structure prediction boundaries are perfectly in agreement with ehl2. The burial in that region is decent using both near and burial. Thu Jun 19 15:32:02 PDT 2008 Firas Khatib looking at Zhang-Server_TS4, I like the N-terminal helices: 1-117 they pack well according to both near and burial and although the ehl2 prediction (which I believe is based on t2k) has coil at 49-50, the t06 dssp- ehl2 predictions have that predicted as helix, which is what 27-58 is here. [KJK: correction---the ehl2 script is the consensus of almost all the secondary structure prediction] I also prefer the 174-222 section of PSI_TS3 over PSI_TS1, it has even better burial and the strand could possibly form with the main sheet in the center. Thu Jun 19 16:04:06 PDT 2008 Firas Khatib I am running VAST on the regions that I like (and it is taking forever for some reason) I created a new directory called vast_searches/ to store the hacked PDB files that I created to run VAST. I rather like the prediction MUFOLD-MD_TS1 The N-terminal region from 1-117 packs pretty well and even the C-terminal end is ok: 156-223. I think if we could start a run using this model and force 138-142 to form a strand with the center sheets, this might be a decent model. (Basically 117-155 is complete garbage) Fri Jun 20 15:06:48 PDT 2008 Firas Khatib Chirag found two decent helices in Zhang-Server_TS5, 1-49, they pack together nicely. Fri Jun 20 16:21:36 PDT 2008 Kevin Karplus try2.costfcn was created to have constraints for the sheets from try1 and some of the helices from PSI_TS3, Zhang-Server_TS4, and MUFUL-MD_TS1. No constraints were added for the packing of the helices, which is really the good part of the pieces of the models. Mon Jun 23 14:54:20 PDT 2008 Firas Khatib so the top scoring server model according to our try2 cost function is BAKER-ROBETTA_TS1. I don't really like how the helices pack (174-222 has better packing in PSI_TS3 in my opinion) but it did attempt to form a third strand in the sheet which I think is promising. Mon Jun 23 17:50:39 PDT 2008 Firas Khatib I am running try3, which is just to try to get that third strand to form with the sheet. It is only reading in BAKER-ROBETTA_TS1 and I manually had to put in sheet constraints for the third strand. (I ran make decoys/score-all.try3.pretty first to make sure my sheet constraints were correct, and they seem ok) Wed Jun 25 14:53:22 PDT 2008 Firas Khatib well the strand was able to form with the main sheet in try3, but it's very short. I think I need to extend the sheet constraints if we think this sheet indeed has 3 strands. The only problem is that is forces the 228-236 helix at the C-terminal end to bunch up against the 175-185 helix if we do this. Thu Jun 26 17:26:31 PDT 2008 Firas Khatib I am running try4 to try to form a better sheet with the 3rd strand. I made the sheet longer in try4.costfcn and increased the weight of that constraint. try4 will read in the top 3 different files scored by try3.costfcn: BAKER-ROBETTA_TS1 T0443.try2-opt3.pdb T0443.try3-opt3.pdb try4 is running on shaw Fri Jun 27 13:13:38 PDT 2008 Firas Khatib dang, that strand paired up worse than try3! I am running try5 with the same strand constraints but removing the helix ones and not reading in the Robetta model. try5 is running on shaw Fri Jun 27 14:25:30 PDT 2008 Firas Khatib try6 will be the same costfcn as try5, but with only T0443.try4-cmoved.pdb as input. try4-cmoved is based on try4 with a chain break inserted at the C-terminal end. try6 is running on lopez Fri Jun 27 19:30:31 PDT 2008 Firas Khatib the try6, try5, try4, try3 and try2 costfcns all agree that T0443.try5-opt3.pdb is the best! And try5-opt3 was able to form a third strand on the sheet! what is odd is that the previous sheet does not look as good as before, even though try5-opt3.sheets is the same as all the other tries (and the new strand is not in the .sheet file, so undertaker doesn't think it formed I guess) Also, try6-opt3 scores better with the rosetta_1.sheet constraint even though that sheet doesn't look better at all: score of -21.2 vs -14.6 in try5-opt3 I am running try7 on shaw, it uses try5-opt3 as input and includes the helix constraints that I had taken out of try5 and try6 Sun Jun 29 09:21:26 PDT 2008 Kevin Karplus With the try7 costfcn, the best models are try7-opt3 try5-opt3 try4-opt3 try3-opt3 try6-opt3 try2-opt3 MQAU1-opt3.repack-nonPC MQAC1-opt3.repack-nonPC try2 costfcn is similar, except that try3 moves before try4 and try2 before try6. Rosetta likes best try2-opt3.gromacs0.repack-nonPC Firas has been making models, but not recording in superimpse-best.under which ones he thinks are currently best. I'll put the try7 ranking in there for now. try7-opt3 and try5-opt3 are essentially the same backbone, so there is no need for both. It is only a little different from BAKER-ROBETTA_TS1, which it is based on---trying to turn the 2-strand sheet into a 3-strand sheet. If we believe this part of the model, then we should try making a subdomain prediction for D120-A248. I think that maybe we should be trying for a 4-strand sheet, but I'm not sure exactly where the 4th strand should be. I started a subdomain prediction for D120-A248. The reason that the try6 run to optimize try4-cmoved didn't do much, I think, is that the try4-cmoved only had one segment moved. It probably got placed back pretty much were it started and so no major rearrangement was done. When fragmenting a model to make it move things around, I think that you need to break things into several fragments in the region where change is desired. I'm going to make a chimera with try7-opt3 for the C-terminal domain and Zhang-Server_TS4 for M1-Q117. I'm also going to try to clean up the sheet a bit. The n_sep and o_sep predictions are for an N163.O-D166.N Hbond, and a W161.O-M169.N Hbond, though there is also support for D169.O-N163.N, W161.O-V168.N, W161.N-(M169.O, V168.O, or Y170.O) Let's try for the simple hairpin with sum 339: SheetConstraint A157 N163 T172 D166 Hbond N163 SheetConstraint L158 R162 P227 R223 hbond A164 Sun Jun 29 10:21:14 PDT 2008 Kevin Karplus I've started an optimization from chimera-zhang4-try7 as try8 on peep. Sun Jun 29 12:51:32 PDT 2008 Kevin Karplus try8-opt3 does not score as well as try7-opt3 or even try5-opt3, largely because of bigger clashes. The D120-A248/try1 run picks out a hairpin at SheetConstraint (T0443)W208 (T0443)K215 (T0443)I225 (T0443)G218 hbond (T0443)K209 1 which does not correspond to anything else we've been looking at, and doesn't even match the subdomain secondary structure prediction very well. I think I'd like to try a different arrangement of strands for try9: SheetConstraint I160 N163 Y170 D167 hbond W161 SheetConstraint Q152 E155 W161 L158 hbond I160 SheetConstraint R223 P227 V168 T172 hbond T172 Sun Jun 29 19:54:31 PDT 2008 Kevin Karplus try9 seems to have gotten a good "sheets" score having modified try5-opt3, though the sheets have not quite formed. To speed things up, since the target is by noon on Jul 1, I'll make a trimmed version of this model and work just on D120-A248 Sun Jun 29 20:08:55 PDT 2008 Kevin Karplus I've turned up the sheet constraints and am running just the second domain as D120-A248/try10. If that succeeds, I'll try superimposing it on try8-opt3 and making another chimera to optimize. Sun Jun 29 21:37:50 PDT 2008 Kevin Karplus the D120-S248/try10 run is a complete mess, because the read-pdb file didn't include the from-try9 model! I'll have to rerun it as D120-S248/try11 Mon Jun 30 11:18:19 PDT 2008 Kevin Karplus D120-A248/try11 makes very little change to the from-try9 model, improving the sheets somewhat in score without making a visible difference. I'll up the weights for the sheets and try again for D120-A248/try12 If that doesn't work, I'll add some distance constraints for CA atoms that are 2 strands apart, to try to flatten out the sheet. Mon Jun 30 14:08:45 PDT 2008 Kevin Karplus D120-A248/try12 didn't change things much. Let's add CA-CA and CB-CB constraints for 2-strands apart: type CA-CA CB-CB parallel 4.84 5.14 antiparallel unbonded 4.43 5.58 antiparallel bonded 5.25 5.02 If we force the sheet to be flat (not quite the normal situation), we get P+ap_un 9.27 10.72 P+ap_bonded 10.09 10.16 ap+ap 9.68 10.60 156< TEAEQL 158> LLIWRN 172< TQYMVDD 227< PIIIRQE So this would pair CA CB E155 T172 9.68 10.6 A154 Q171 9.68 10.6 E153 Y170 9.68 10.6 Q152 M169 9.68 10.6 L151 V168 9.68 10.6 L158 P227 9.27 10.72 L159 I226 10.09 10.16 I160 I225 9.27 10.72 W161 I224 10.09 10.16 R162 R223 9.27 10.72 N163 Q222 10.09 10.16 Mon Jun 30 16:49:31 PDT 2008 Kevin Karplus Even D120-A248/try13 doesn't manage to make this into a sheet. I'll make a chimera of try13 with try8, but it will probably be a low-ranked model. Mon Jun 30 21:27:00 PDT 2008 Kevin Karplus try14 is pretty junky, but I'm out of time and can't do much more exploration. I'll submit ReadConformPDB T0443.try9-opt3.pdb.gz # < try5-opt3 < try4-opt3 < BAKER-ROBETTA_TS1 ReadConformPDB T0443.try8-opt3.pdb.gz # < chimera-zhang4-try7 ReadConformPDB T0443.MQAU1-opt3.repack-nonPC.pdb # < SAM-T08-server_TS1 ReadConformPDB T0443.try2-opt3.gromacs0.repack-nonPC.pdb # < MQAC1-opt3 < SAM-T08-server_TS1 ReadConformPDB T0443.try14-opt3.pdb # < chimera-try8-try13