Wed Jun 9 10:20:51 PDT 2004 T0201 DUE 5 Aug 2004 Wed Jun 9 12:58:17 PDT 2004 Kevin Karplus This looks like it will be a new-fold prediction. There are probably 2-3 helices and a 3- or 4-strand sheet. The strand-helix-strand alternation suggests a parallel sheet, but the helices are rather long, and the str2 neural net is predicting antiparallel strands (but it usually prefers to predict that). Mutual information doesn't help here---there are no significant pairs, becasue the t2k alignment found no homologs. Perhaps we need to run on the t04 alignment instead. I'll try make AL_METHOD=t04 mutual_info This does give some pairs to try. I'll try putting them in for try2.costfcn. Oops---just noticed a bug in try1.costfcn---the dssp-ehl constraints were not properly included. I might as well switch to the t04 constraints for secondary structure also, even though we don't have the new networks yet. In the t04 alignment, I see highly conserved H16 and C23. I wonder if they coordinate a metal ion? If so, we'd want to add some more constraints. Wed Jun 9 15:28:25 PDT 2004 Kevin Karplus I'd like to add a parallel strand constraint for L52-I85 V54-V87 I don't know which way they should Hbond, so let's just try CB constraints around 5 Ang. Wed Jun 9 17:28:37 PDT 2004 Some good things appear in try3, there are large breaks, and the parallel strands I wanted didn't form. For try4, I'll add some new constraints---trying to get the conserved residues to cluster (they already almost are). H16.ND1 C23.SG I2.CD1 L77.CD1 F81.CZ Then I'll try adding anti-parallel connections (with Hbonds): F11-V87 V13-I85 Wed Jun 9 19:53:44 PDT 2004 Kevin Karplus try4 forms a 3-strand sheet, but almost makes a knot. Maybe K44-K47 and Y51-F56 form a hairpin? A43-V54 I45-L52 Let's try that for try5. Thu Jun 10 15:12:48 PDT 2004 Kevin Karplus try5 is hideous. All the beta sheets are gone. I'll try picking up hbonds from try4 and from alignments and running try6. I'll also reduce the break penalty slightly, so that things have more freedom to move around before the loops get closed. Fri Jun 11 14:35:44 PDT 2004 Kevin Karplus try6 is pretty bad also. There are beta sheets, but nothing else looks reasonable. I'm not sure what to do to fix this so that it packs more reasonably. Take out the constraints, increase beta hbond weights, and rebuild from many starting points? This looks like it is small enough that it should be feasible to construct, but I'm not having any good ideas. Sat Jun 12 21:03:42 PDT 2004 Kevin Karplus try7 packs the helices nicely against each other, but only makes one hairpin, and winds one of the predicted strands into a helix. I'm NOT making much progress here! Thursday June 17 14:27 2004 Bret Barnes So I made a few changes to try8. First I added some constraints for another beta strand from K44 to K46. All the seondary structure predictions showed evidence for a strand here. After reviewing all the current predicted structures it seems that this strand probably h-bonds with strand Y51 - F56 and allows the helices to wrap around eachother. I also removed the constraints from T0201.t04-thin30 because they didn't seem to work with the constraints from thin50 and thin62. Thin50 and thin62 are more related in sequenes and therefore probably reflect the similar constraints. Since I suspect that the new strand I added actually binds to strand Y51 - F56 I removed the constraints to make Y51 - F56 bind to the last strand I85 - I90. After looking at the nice anit-parallel beta sheet in try4 it seems more reasonable that I85 I90 h-bonds with F10 - T7 strand. Also when you review the CB-burial-14-7 predictions these beta-sheets should be exposed and try to bury the helices, which they would if they bound this way. It also reflects the anti-parallel predictions made by the str2 predictions. I added H-bonding between the first strand (K3 - T7) and the second strand (F10 - T7) since this appeared in almost every structure. One possible change that might be worth to try is to create a disulfide bridge between the only two cystines in the structure. The main reason for this would be because this protein was purified from Thermotoga maritima which is a single celled organisms that lives near thermovents off of Italy. Proteins that operate under high temperature usually use disfulfied bridges to keep their structure. The one cystine is highly conserved, but the second one is not. This could be because in homologous proteins the 2nd cystine is located in diferent positions to create different structures. In addition small proteins usually use things like disulfide bridges or electrostatic interactions to hold their structure together since they can't rely on many hydrophobic and hydrophilic interactions. Also, if you look at the surround residues next to the second cystine (C62) you notice many other catalytic residues like glutamic acid that could interact with the histidine next to the first cystine (C23). There is also an aspartic acid (D19), lysine (K20) and two serine (S25, s27) near the highly conserved histidine (H16). This could be totally wrong, but if try8 happens to get thses cystines close enough it could be a possble structure for try9. Thursday June 17 19:19 2004 Bret Barnes So, try8 looked pretty bad. I think this is because I used the previous models in the try. This time I removed those and did a new fresh run. I also included one of the thin30 constraints that I removed before to try to make the protein more compact since undertaker is bad at that. I removed the thin50 constraints and replaced them with L37 - L67 constraint and Q31 - M70 constraint. These constraints seemed to represent the helix packing better when looking at the burial plots on rasmol with try3 (which has the best helix packing). I also changed my previous constraints that tried to bond strand 2 with strand 5 with Hbond constraints that were there previously. Finally I increased the super_iter to 15. Hopefully this will produce a more compact structure with the 2ndary structures in the correct place. Fri Jun 18 11:04:05 PDT 2004 Kevin Karplus Changed Makefile so that t04 (which has more sequences) is used as the "preferred" alignment method. Bret, your Hbonds in try9 for I2 and K3 make no sense---the hbonds should be every other residue. To line up the anti-parallel strands at the beginning, I'd recommend T7-F10 T5-E12 K3-T14 From bbarnes@ucsc.edu Sat Jun 19 18:10:48 2004 Date: Sat, 19 Jun 2004 18:10:24 -0700 (PDT) From: Bret D Barnes To: Kevin Karplus cc: katzman@ieee.org, sol@soe.ucsc.edu, ggshack@soe.ucsc.edu, learithe@soe.ucsc.edu, martina@soe.ucsc.edu Subject: Re: deadlines Kevin, I have been working with T0201, its a small protein. I think the best thing we can do is pack the helices together first and work from there. If you have time I want you to look at what I have written in the README file. Try8 and try9 do not look too good. If you have any suggestions that would be really helpful. Otherwise, I'm going to put a constraint to pack the helices together and work from there. -Bret -------------------------------------------------- Sat Jun 19 19:35:30 PDT 2004 Kevin Karplus I started looking at some of the predictions in rather arbitrary order. Given that there is a HIS in the immediate neighborhood, I'm inclined to believe more in a metal-binding site than in a disulfide. H16 and C23 are conserved residues, which makes a metal-binding site even more appealing. H32 and C62 may or may not be involved. I suspect that I2, L77 and F81 (conserved hydrophobics) may be near each other in the core. try9-opt2 Looks terrible. Very open and rather gappy. try8-opt2 Forms one hairpin but otherwise looks poor. try7-opt2 makes almost everything into helices and puts stuff predicted to be buried on the outside. try6-opt2 forms a helix but does quite form even a 3-strand sheet. Not compact. try5-opt2 has nothing--gets 2ry prediction all wrong. Without looking at my earlier runs (so as not to bias my opinions) I put together conjectured hairpins, the conjectured metal-binding site, and the conserved hydrophobics in the t04 multiple alignment. I did not use any Hbonds (though there are some guesses commented out in try10.costfcn). With this cost function, try8-opt2 scores the best (probably because of bringing the cys close). I put back the commented-out TryAllAlign commands in try10.under, though I don't know how much they will help. If this run comes up with anything, we should grab the hbonds and freeze them, then try adding new constraints to bring the hairpins together. The strand-helix-strand motif is usually a strong signal for a parallel right-handed connection (Z with the diagonal bar of the Z being the helix in front of the two strands). I may be trying too hard to form the metal-binding site. It may be better just to try the conserved residue clusters: H16 and C23 together and I2, L77, and F81 together. ------------------------------ From karplus@soe.ucsc.edu Sat Jun 19 20:38:24 2004 Date: Sat, 19 Jun 2004 20:38:22 -0700 From: Kevin Karplus To: bbarnes@ucsc.edu CC: katzman@ieee.org, sol@soe.ucsc.edu, ggshack@soe.ucsc.edu, learithe@soe.ucsc.edu, martina@soe.ucsc.edu, karplus@soe.ucsc.edu In-reply-to: (message from Bret D Barnes on Sat, 19 Jun 2004 18:10:24 -0700 (PDT)) Subject: Re: deadlines I'm not sure what T0201 should look like. I made some remarks in the README file and tried starting a new run (try10) to see what came out of it. I may do another run tonight or tomorrow morning. This one requires patience and some luck in building the beta sheet. I'm pretty sure that this is a single 5-strand sheet, but I don't know whether both helices are on the same side or opposite sides. I favor same side slightly, but couldn't justify my preference. Kevin ------------------------------------------------------------ Sun Jun 20 08:27:05 PDT 2004 Kevin Karplus Well, try10 is no great model. H16, C23, and H32 are clustering reasonably, but C62 seems to be moving the helix it is on to a strange place. The conserved hydrophobics are not clustering. The hairpins that are formed seem plausible, and most of the residues predicted to be buried are, escept for A24-L29, which are predicted to be very buried. Strand Q84-I90 is not paired with anything. Looking at the Hbonds formed in all models so far for popular pairings: M1 E12 or A17 K3 F10,E12,T14, or G15 T5 E12 T7 F10 F11 V87 V13 I85 V26 A34? N35 K44? K38 A43? K44 K55? K46 K53 E48 Y51 So let's continue trying K3-T14, T5-E12, T7-F10 and K44-K55, K46-K53, E48-Y51 using Hbond specifications, and let's try removing the C62 constraint, adding the D91-E48 constraint from t04-thin35.mi.constraints. I'll also try forming a parallel connection: L52-I85, K53-R86, V54-V87, K55-E88, F56-V89 Sun Jun 20 11:37:16 PDT 2004 Kevin Karplus try11 groups several of things I had hoped to see together, but still doesn't form a sheet. I'm now thinking that there may be 5 strands arranged like 4> 3< 1> 2< 5> Nope--that crosses the helices, unless one is a left-handed crossing. We can make the two helices parallel by doing 4> 3< 5> 1> 2< The 1-2 and 3-4 hairpins are probably right, but we need to try to line up strands 1 and 5 and 3 and 5. In strand 1 we can hbond to I2, V4, V6, and on strand 3 A43, I45, and K47. 50< gsekkikakk 84> qirvevid 1> mikvtv i'll try this set of constraints as try12, before turning the optimization back over to the students. Sun Jun 20 18:31:36 PDT 2004 Kevin Karplus Try12 looks like it is trying to form the sheet, but I am having trouble with the phase of the Hbonds, so that strand 5 is not behaving. Also the helices seem to be passing on the exposed side of the sheet. not the buried side. Maybe I need to add some CB-CB constraints for strands 1 and 3 to push them apart enough for strand5 to sneak in. How about V6 about 10.5 from A43 and V4 10.5 from I45. We could have V87 between I2 and K47 Hmm---may it would be better to hand this back to Bret for awhile. Tuesday June 22 14:04 2004 Bret Barnes In try13 I tried to force strand 5 to fit between strand 1 and stand 3. I increased the sheet constraint for strand 5 by two for the str2 secondary structure prediction. Perhaps this might cause the sheet to form better. I also added Hbonds between sheet 1 and 5. StrandConstraint R86 I90 0.824 # previously it was .624 Hbond V4.N E88.O 1.0 Hbond V4.O E88.N 1.0 Hbond V6.N I90.O 1.0 Hbond V6.O I90.N 1.0 I also tryied to cluster H16, H32, C23, and C62 together to form a metal binding site. New constraints shown bellow. Constraint C23.SG C62.SG 3 4.5 7 1.0 Constraint H16.NE2 C62.SG 3 4.5 7 1.0 Constraint C62.SG H32.NE2 3 4.5 7 0.5 If this doen't help wedge strand 5 into position, then I'll try using Kevin's suggestion about pushing strand 1 and 3 away from eachother with CB-CB constraints. (ie. V6 10.5 away from A43 and V4 10.5 away from I45). Thursday June 24 17:39 2004 For try14 I did a more indepth analysis of the beta sheets. I tried to find the best agreement between try12's and try13's beta h-bonds. From that data and by looking at the positioning of hydrophobic and hydrophilic residues on parallel and anti-parallel sheets I concluded with the following alignment between the beta strands. Alignment Diagram: Strand 2: 10 FFEVTG 15 | | | Strand 1: 9 SNTVTVKI 2 | | | Strand 5: 92 NDIVEVRI 85 || Strand 3: 44 KIKKE 48 | | | Strand 4: 55 KVKLY 51 Hydrophobicity Diagram (Based on Amino Acid Hydrophocity Table) Key: B - Hydrophilic b - Amphipathic r - Hydrophobic * - Glycine Strand 2: 10 rrBrb* --> (Anti-Par) | | | Strand 1: 9 BBBrBrBr <-- (Parallel) | | | Strand 5: 92 BBrrBrBr <-- (Anti-Par) || Strand 3: 42 rbbBbBbbBB --> (Anti-Par) | | | Strand4: 57 rBrBrBrB <-- Parallel Strands like 1 and 5 should have matching hydrophoics and matching hydrophilics. This would put all the hydrophobics on the same side and all the hydrophililcs on the other side. Anti-parallel strands should have mismatching hydrophobic and hydrophilic residues. In addition to adding these constraints (most of which were already included) I removed the previous zinc finger constrands and conserved hydrophobics constraint in hopes that I could get undertaker to really want to create the beta sheet. Wed Jun 23 14:02 2004 Bret Barnes This time I tried including the zinc finger constraints and clustered the conserved hydrophobics. I used the same sheet constraints as try14. Tue Jul 6 15:12 2004 Bret Barnes I made the weights for creating the zinc finger even. I also changed the strand constraints into strandConstraints so that the H-bonding is set up correctly. Hopefuly this will complete the beta-sheet. Fri Jul 09 16:03 2004 Bret Barnes So I removed the conserved hydrophobic constraints. I also removed the C62 constraints to try to form a zinc finger. Instead I tried to add D19 or S73 to the metal binding site. I also increased the sheet constraints and increased the break constraint. We want to pack both of the helicies on the hydrophobic side of our sheet since the sheet is ampapathic and the helicies have hydrophobic sides. Sat Jul 10 12:34 2004 Bret Barnes I just re-ran try17, but lowered the break constraint. Try17 didn't form the sheets at all. Try18 formed the the hairpin strands (sheets 1-2 and 3-4). Strand 5 is in the right spot, but is not a strand yet. The helicies look much better right now. They are both on the right side of the sheet and are looking like the model that we had for this protein. Tues July 13 15:05 2004 Bret Barnes I'm going to use the 2 sheet constraints that try18 produced in the beta-hbonds file (for strands 1-2 and strands 3-4). I'm also going to bumb up the weights for these constraints. I'm going to cluster HIS 16, HIS 32, CYS 23, and ASP 19 to try to form a metal binding group. Also I believe that by puting SER 73 near HIS 32 and CYS 23 the helicies were forced to be in the righ spot. This time I'm going to just put SER 73 next to HIS 32, and add a LEU 37 and VAL 66 constraint to tighten the helicies even more and try to slide them into the right spot. If this run looks good I might add some constraints to force the helicies to pack tighter against the sheet. Also, if I can't get the 5th sheet to form I might have to fudge around with that. Wed Jul 14 15:44 2004 Bret Barnes Ok, so try19 and try20 look much worse than try18. I guess I'm going to go back to try18 cost function and work from there. I'll probably leave the strand constraints alone, and try to get the helicies together and what might be a metal binding site. Thu Jul 15 14:57:21 PDT 2004 Kevin Karplus With the try21 costfcn, the best-scoring is try16-opt2. I've moved the CreatePredAlphaCost into the try22.costfcn and no-sheets.costfcn out of score-all.under and try22.under. I've written rasmol scripts "metal" to highlight the putative metal-binding site, and "strands" to define s1 through s5. I've modified try22 to have our current best guesses about the sheet topology and neighbor constraints---plus some conjectures about how to pack the helices against the sheets. Now try15 and try18 are preferred to try16. I'll do another run from the alignments, but pretty soon we'll have enough models of comparable value that we can start optimizing from the existing models. Mon Jul 19 15:42 2004 Bret Barnes Ok, so this time I'm going back to try18 cost function. The first changes I made was to raise the constraint on the 5th strand. I removed all the constraints on HIS32 since it was too far away from the possible metal binding site. I still clustered D19 in the area since it is fairly conserved and has many other possible metal binding ions in its area. Next I added the first and third highest RR constraints from the 134 set. These predictions agreed with the try18's structure, so I included them and bumbed up their weights 1.0 each. I think these should help the helicies pack together. L77 and F81 are two of the conserved hydrophobics that are spacially close in try18's structure so I threw them into the mix. I kept the sheet constraints for sheet 1-2 and 3-4, but changed the sheet topology to: 1< 2> 5< 4> 3< This seems more reasonable than the previouse topology: 2> 1< 5< 3> 4< because it puts the hydrophobic sides of the sheet on the same side as the hydrophobic parts of the helicies. I didn't know the exact hbond alignment of the sheets so I just used CB-CB constraints in hope to get some sort of sheet starting to form. The new constraints were sheet 2 (10,14) aligned with sheet 5 (90,86) and sheet 4 (52,56) aligned with sheet 5 (86,90). I also noticed that the ReadFragmentAlignment T0201.t2k.many.frag was reading t2k which is a very bad alginment set for this protein. So I changed it to: ReadFragmentAlignment T0201.t04.many.frag Tues Jul 20 12:21 2004 Bret Barnes Ok, so the sheet topology of try23 is almost there. Strands 1,2 and 5 are in the correct order. Strands 3 and 4 still need to be swaped, but they are getting close. The RR constraints that I am using to pack the helicies are mostlikely right, but I think they are causing the helicies to break. So, I am going to lower their weights. Also, the D19 constraints to the metal ion binding site could be distrupting the helicies too, so I am going to remove those. Since the 4-5 sheet is not forming yet I'm going to increase the weights on the CB-CB constraints. Running try24 on meow. Wed jul 21 11:19 2004 Bret Barnes So, try24 started to put strand 4 in the right spot, better than try23, but now the helicies are in the wrong position. I'm going to add back the D19 constraint for the metal ion binding site because I think it might help to keep the first helix next to strands 1 and 2. Also, I'm just going to raise the hell out of the sheet constraints to get the sheet to form. If this does not get the sheet to form I'm going to take try18 (which looks really good, but has the wrong strand order 21534 instead of 12543) and the best of the new trys (with the 12543 strand order) and use DALI or VAST to see if there is any similar structures. I'm also adding the ReadFragmentAlignment T0201.t2k.many.frag, before I was just using the t04 on the new ones. I uncommented the PrintTemplateAtoms Template.atoms. Fri 23 16:19 2004 Bret Barnes On try26 I only used sheet constraints to just get the sheet to from, but that didn't work too well. I'm now going to use the bonus constraint to try to put the helicies in place. I hope that this will get us closer. I'm pretty sure what the structre is, but now its a process of finding what constraints to use to form it. running try27 on cluck Fri Jul 23 22:24:39 PDT 2004 Kevin Karplus I noticed that the Template.atoms file was being written to decoys/ rather than the main directory. I gzipped it and moved it to the main directory, where it would do some good. I made an "unconstrained" costfcn, to see which models looked reasonable---it likes try7-opt2 best, then try2-opt2, try1-opt2, try17-opt2, try10-opt2, try23-opt2, robetta-model3, ... I set up the MANUAL_TOP_HITS line in Makefile, and ran "make extra_alignments" to make the alignments for all the top hits. Then I did "make all-align.a2m.gz" Adding in strand and helix constraints to unconstrained.costfcn to make try29.costfcn still has try7 at the top, then try2, try1, robetta-model3, ... Adding in the putative metal-binding side (C23,D19,H16,H32) changes the order to try10, try21, try23, try7, try15, try19, ... Adding in the two hairpins from try10-opt2 (but not the part before the bulge), makes the order try10, try23, try18, try15, ... The best way I see to get the helices on the same side of the sheet (which seems the more likely configuration) is to have the strand order s1 ^v s2 ^v s5 ^v s3 ^v s4 or s2 ^v s1 || s5 ^v s3 ^v s4. But s3 looks like an edge strand---hmm, maybe s5 ^v s2 ^v s1 || s4 ^v s3 ? That looks fairly compatible with the try10-opt2 model, so let's try adding constraints for it. With those constraints added, try19 scores best, followed by try16, try15, try25, try26, try12, try24, try27, try25, ... Sat Jul 24 14:12:35 PDT 2004 Kevin Karplus With the unconstrained costfcn, best are now: try7, try2, try1, try17, try29, try10, try23, ... With try29.costfcn, the best are try29, try19, try16, try15, try28, try25, ... try28 looks pretty bad---I suspect inconsistent sheet constraints. try27 also has problems that look hard to fix. In try29, the two helices pack against each other ok, but the first hairpin looks pretty messed up. I don't really see any changes I want to make to the sheet constraints, though. I could try adding some constraints on side chains to keep the strands properly oriented, and increase the weight of constraints to try pusshing harder. Sat Jul 24 21:09:57 PDT 2004 Kevin Karplus Try30 looks a lot like try29---I think it's gotten stuck. I'll have to think about what to do on this one. Sun Jul 25 14:40:23 PDT 2004 Kevin Karplus It looks like Bret is doing something, but he hasn't shared his thoughts in the README file. I started a try34, which attempts to close up the gaps in try30 and clean up the hbond constraints. Sun Jul 25 16:33:37 PDT 2004 George Shackelford I'm going to try doing some constraints built around the RR constraints. There is a hint that one of the helices needs to turn around. TRY35 running on crow. Sun Jul 25 19:18:15 PDT 2004 Kevin Karplus With the unconstrained cost function, the current best models are try7, try30, try34-opt1, try2, try1, try17, try29. Try34 made only tiny changes to try30, and is not significantly improved. Perhaps I need another run from the alignments, to get out of the rut. Try36 (running on caw) is such a run, using exactly the same score function as for try34. There is no point to provide a model from alignment for this target, so model5 should be try1-opt2. That leaves us 4 models to choose. Interestingly, try7 is a model that has one helix on each side. I did not explore those topologies much, since predicted burial seemed to indicate that one face of the sheet was exposed. Sun Jul 25 19:50:24 PDT 2004 Kevin Karplus try34 just finished, and now scores best with the unconstrained cost fcn. Right now I favor try34 try19?? try16??? try10??? try7-opt2.repack-nonPC which rosetta likes best try1 Sun Jul 25 22:15:49 PDT 2004 George Shackelford I like try-opt2.repack as well. But I don't think it's right. I'm going to get 'crazy' and try something tough. Crank up the RR constraints and break. See what I get. Working on try37. I'm turning off SCWRL as well. Sun Jul 25 22:44:41 PDT 2004 Kevin Karplus George left out the most critical piece of information: WHICH try did he like as well? The unconstrained costfcn now scores try35-opt1 best, order try35-opt1, try34, try7, try30, try36, try2, try1, ... I'm hard pressed to see much distinction between try34 and try35. This is stuck in a rut---if we want to see something different, we need to back way off and start over from alignments. It looks like try35 died or was killed at 22:25. No great loss, since it was doing minor polishing on try34. Sun Jul 25 22:53:54 PDT 2004 George Shackelford Ugh! That should have said: try7-opt2.repack-nonPC. I killed the try35. Yes, it did not look different at all. Also it was taking longer due to using SCWRL. Try37 is now running on croak. Mon Jul 26 07:16:20 PDT 2004 Kevin Karplus I don't like try37 at all---it has wound all but 2 of the strands into helices. The try37 cost function scores try35-opt1 best, then try37, try34, try30, ... The unconstrained cost function also liked try37 best! The order is try37, try35-opt1, try34, try7, try30, try36, try2, try1, ... Mon Jul 26 07:37:38 PDT 2004 George Shackelford Try37 looks interesting. Think I'll drop the ehl2 constraints down a bit and see what happens. Keep the rest where they are. Time to get REALLY crazy... running try38 on croak. Mon Jul 26 10:45:33 PDT 2004 George Shackelford Hmm. pretty, but I think I'm going to crank up the RR's again! REALLY REALLY crazy! Hope I can get on croak For TRY39 I've crank up the 'constraint' to 40, and dropped down the metal binding. Now running on croak. Mon Jul 26 13:56:26 PDT 2004 Kevin Karplus George will categorize the models to see if we have some different topologies. Mon Jul 26 16:13:03 PDT 2004 George Shackelford Topologies of the models have been charted. Only a couple are valid. Starting with 36, I'm doing TRY40 using topology 5v2^1v4v3^ Mon Jul 26 17:28:35 PDT 2004 Kevin Karplus It appears that George left the results of his categorizing in "tries.table", though he neglected to say so. George likes try36 5 ^v 2 ^v 1 || 4 ^v 3 try25 1 ^v 2 ^v 5 || 4 ^v 3 I'll add try35-1 5 ^v 2 ^v 1 || 4 ^v 3 try18 2 ^v 1 || 5 ^v 3 ^v 4 We need try1 Mon Jul 26 21:23:12 PDT 2004 George Shackelford Try40 is much better but had some problems. S5 curled into a helix, S1 and S2 needed to be flipped over so the hyrophobes faced down, and S1 is || to S4 not ^v. I set StrandConstraint for S5 to 20, changed the hbond linking S1,S2 to flip it, and uncommented the original S1 || S2 constraint. I also adjusted the range for S2 in that parallel match to match up the 'near' values better. TRY41 now running on 'peep' I also looked at the RR constraint matchups. They really aren't matching up; I am going to study it to see if I can find a topology that can work with them. I think I'll try an experiment. Turn off the topological sheet constraints and turn up the RR constraints. See what happens. Starting TRY42 now running on 'ribbit' Mon Jul 26 23:27:18 PDT 2004 George Shackelford Got try41 back. Had an error in bonding S2 to S5. That twisted S2 causing S1 to not line up right. Then S2 didn't bond to S4, but I don't know why. Fixed the bond between S2 and S5 and shifted the match by 2 residues so S1 could meet up with S4 better. Now TRY42 running on 'peep' Tue Jul 27 10:30:02 PDT 2004 George Shackelford What is going on?? S5,S2,S1 bonded but S4,S3 broke. Perhaps the length of H1 is forcing it to break. I'm going to turn off the S1||S4 constraint and relax the basic sheet constraints from 20 to 15. Also relaxing the RR constraints. Now let's see where the sheets fall, then rejoin them. I'm also concerned with the hydrophobic pattern on H1. It seems to cause H1 to break in two. TRY44 going on 'peep' Tue Jul 27 17:10:33 PDT 2004 George Shackelford try44 looks a bit like a sheet between two helices. I am going back to try42 which has the two helices together and I am going to try a different topo: S5||S1^vS2^vS4^vS3. Try42 seems to hint in that direction, all it may need is to join S1,S5 and S2,S4. Doing so for TRY45 running on 'ribbit' Wed Jul 28 10:48:22 PDT 2004 George Shackelford try45 had line up and hbond problems. Thought I had fixed them in try46, but I didn't. Now I'm running TRY47 on 'peep' and I believe I have it all in place. I also used some hints from the RR constraints as to lining up 6 with 87. Wed Jul 28 13:01:49 PDT 2004 George Shackelford Results of try47. I 'fixed' some hbonds that did not need fixing. I have unfixed them. TRY48 now running on peep. "Dave, my mind is going. I can feel it." - HAL Wed Jul 28 15:29:50 PDT 2004 George Shackelford The result of try48 is that, yes I DID need to fix those hbonds. At least the structure is doing well otherwise. For TRY49, I lined up 11 with 52 because 52 lines up with 45 which has an RR constraint with 11. It's a guess along with lining up 6 with 87 (see above). TRY49 running on peep. Thu Jul 29 00:56:44 PDT 2004 George Shackelford Try49 still has problems. Apparently the niceness part, the S4,S3 sheet had a bad hbond. I fixed that and I'm trying another rerun. I also double checked the other sheet constraints and tighten the ranges slightly. We'll see the results in TRY50 (sigh) running on peep. Fri Jul 30 14:08:23 PDT 2004 George Shackelford Notes I had made yesterday concerning trys 51,52 and 53 apparently were lost. I must have kept my editor open and lost stuff when the system crashed (again). Basically the efforts kept blowning up. Even the corrected S4,S3 hbond made things worse. I am going to take try1.under and the current try53.costfcn and setup try54 just in case something went wrong in parameter changes to the *.under series. Desparate times call for desparate measures. Running TRY54 on peep. Reviewed what George has been doing and I like the 5 || 1 ^v 2 ^v 4 ^v 3 topology. Try55 has that topology. I also added some long distance sheet constraints. By that I mean distance constraints for s5 and s2, s1 and s4, and s2 and s3. This might help form the sheet by restricting the strands movement. However I made all these bonus constraints so they would not mess up the other constraints. I also added another possible ion binding site canadite that also might help the structural integrity of the protein. I added two distance constraints putting E78 close to H16 and C23, both as bonus constraints. Another thing I just noticed is that in the PDB there is a small protein that has the exact topology as try18, that is the s2 ^v s1 || s5 v^ s3 ^v s4. It is 81 residues long (as opposed to 94 like T0201) and it looks like a possible canadite to do structure-structure comparisions with, or even some sort of threading procedure. Its PDB ID is: 1GUA chain B. Its part of a dimer (the other half of the dimer is much bigger; chain A). Oops, I totally forgot to put the beta carbons on my new long distance sheet constraints. I'll fix that and kill the run. Sat July 31 14:12 2004 Bret Barnes Try55 is in the correct topology, but as normal the sheets are not forming. I'll try removing my long distance sheet constraints and see if they were messing it up. Another thing worth noticing is that SER 83 is very close to our metal ion binding site. Perhaps its worth it to give a bonus for putting that near the binding site. Sun Aug 1 11:01:30 2004 George Shackelford I'm going to try a different topology. It seems a bit odd, but it might work: S1^vS2^vS5^vS3^vS4 At least everything is anti-p. I'll try to do it using basic constraints rather than arbitrary sheet constraints so undertaker can adjust as necessary. TRY56 Sun Aug 1 17:42 2004 Bret Barnes I like how try56 came out the sheet is almost formed. I'm just going to raise the constraints for sheet 1-5 and see if that doesn't finish it. I also want to try using 1GUA chain B as a model for a different topology. Mon Aug 2 13:42 2004 Bret Barnes Try58 does not look as good as try56, so I went back to try56 and included the strand and helix constraints, along with the sheet constraints. Hopefully this will get closer to the correct structure. running try59 on woof Ok, so getting back to the try18 topology (ie. s2 ^v s1 || s5 v^ s3 ^v s4) I'm doing a new run very similar to try18. If this looks better than try18 I'll use it to do a structural alignment to 1gua chain b and see what I can do from there. Try60 running on baa If both try59 and try60 look good we will have two possible topologies to sumbit. Mon Aug 2 16:28:17 PDT 2004 George Shackelford I'm going to fix try56. Now that I have decided on the bonds, I'm forcing some sheets very strongly. It will probably break everything... TRY61 on peep. Doing a new try based on the new topo. TRY62 running on peep(?) Try61 had some errors. The specs for a couple of the sheets were completely wrong. Redoing as TRY63 on peep(?). Tue Aug 3 13:49:30 PDT 2004 George Shackelford Try63 blew up bad! Try62 could only get three strands into a sheet! I'm going to review what we have and decide on what to submit. This beast has taken way too much time! Tue 3 18:27 2004 Bret Barnes I'm going to try one more attempt to get the sheet to form. Basically I'm going to give a bonus constraint for sheet 1 || 5 to form, along with the normal constraint. try64 running on woof. Wed Aug 4 12:49 2004 Bret Barnes Try64 puts strand 5 close to strand 1, but doesn't form the sheet. It also puts strand 2 at the wrong angle to be next to strand 4. I'm going to try to add a bonus constraint to get strand 2 and 4 next to eachother and also raise the strand constraint for strand 5. runnin65 on woof. Just as an experiment, I want to see what undertaker does without helix and strand constraints. So I'm going to copy try56 constfunction and under files to try66 and remove the helix and strand constraints. This is more of a learning experiment than a try at the structure. Running try66 on baa. Wed Aug 4 21:18:31 PDT 2004 George Shackelford Unless Bret comes up with a really good result, these are the best I have to submit (first is best): T0201.try56-opt1.pdb s5||s1^vs2^vs4^vs3 T0201.try43-opt2.pdb s5^vs2^vs1||s4^vs3 T0201.try60-opt2.pdb s1^vs2^vs4^vs3^vs5 T0201.try18-opt2.pdb s2^vs1||s5^vs3^vs4 T0201.try1-opt2.pdb automated decoy ? s5^vs2^vs1||s4^vs3 At least we have a coverage of different Topos. I kind of like try60 though it scores badly. Thu Nov 18 22:13:27 PST 2004 Martina Koeva Based on the smooth gdt score results: best sam-t04 48.3157 (try41-opt1) best submit 40.1299 (this was model 1) model1 40.1299 auto 24.9910 align 19.0721 robetta best 35.9514 (robetta model6) robetta1 26.1437 Fri Nov 19 13:38:04 PST 2004 Kevin Karplus The correct structure is a single sheet with s5 || s1 ^v s2 ^v s4 ^v s3, which is what we submitted for model1, though we did not manage to get all the hbonds to form correctly. George introduced this topology deliberately on try45, but it seems to have been forming in try41 already. Good work, George!