Tue Jul 9 17:00:11 PDT 2002 9 July 2002 Kevin Karplus No really strong hits, best seems to be 1efcA, but other, very different domains also score almost as well. This is probably an ab initio target. 10 July 2002 Kevin Karplus The top few hits give fragmentary alignments, not full folds. The try1-opt.pdb prediction has pretty good helices, but the beta strands are not joined into a sheet. It looks like 35-42 and 46-52 should form a hairpin, so we might try adding some constraints to encourage that. Predicted alpha values, once implemented, should help straighten some of the strands that are coiling up. 35 40 | | VEAVIELNEK |||| ) SETTIKITEN | | 50 45 We could constrain the CBs to be 5.12 Angstroms apart across the hairpin: I39 I50 E40 K49 L41 I48 N42 T47 On try2, with the constraints added, there is a beginning of a hairpin between E43 and E46, but the sequences continue to diverge after that. Perhaps we should also try to make a hairpin out of V121-Q124 and Q129-V132 VQTQIQ |||| ) VRVQEG V121 V132 Q122 R131 T123 V130 Q124 Q129 11 July 2002 Kevin Karplus On try3, the later hairpin 121-132 looks nicely formed, but it seems to have flipped one of the strands, with R131 hbonded to T123 instead of Q122. The hairpin centered at 44-45 is also starting to form. It seems pretty clear that the beta strand ending around L98,K99 is paired with the strand ending around Q124. Perhaps the pairing is LYSKEIKLKQ ||| GIETEMAKKITKLVKDSKIKVQTQIQGEQVRVTGKSRDDLQAVIQLVKSAELGQPFQFNNFRD I96 I119 K97 K120 L98 V121 On try4, the beta sheets have still not formed, but there is somewhat more evidence of them forming. 12 Aug 2002 kevin Karplus Reran make with new template library---no changes to top hits. Modernized define-score.script and undertaker.script and reran undertaker. 17 Aug 2002 Kevin Karplus Current best score is try5-opt, but it does not look very good. It may be worth trying to steal pieces of the robetta models---perhaps by starting from them and re-optimizing. The robetta models do not score well with the constraints used in try4 and try5, so let's rescore without constraints and see where they fall. They are not very compact, but the alpha values look good, and robetta4 (which scores best with the unconstrained score function) actually looks like it is about to form a sheet. Before trying to steal the robetta models though, let's do another pure undertaker run using a larger set of alignments. Tomorrow will be soon enough to start from the robetta models. 18 Aug 2002 Kevin Karplus try6-opt scores almost, but not quite as well as try5-opt. Note that try6 was built without constraints, but has about as much sheet structure as try5-opt (that is, a pair of strands that come close---try5 actually forms a 3-strand sheet, but one strand is not predicted to be strand). Perhaps we should try a run starting from the robetta models now, and see how they do. 18 Aug 2002 Kevin Karplus try7-opt-scwrl is new best score. It seems to be based on robetta4, and is still rather molten, though it has a sheet that it got from robetta4. The sheet in try7 is certainly much nicer than the junk in try6, and matches 2ry prediction better than try5's sheet does. I think I'll need help trying to get constraints to pull this together. 20 Aug 2002 Rachel Karchin strands are still crinkled and broken. I'm trying to come up with a good topology for this structure, based on the STR prediction, and looking at hydrophobic packing between helices and strands. rough definition of predicted secondary structures T11-R28 Helix-1 E36-N42 Strand-1 E46-T52 Strand-2 D55-I70 Helix-2 L91-L98 Strand-3 T104-D116 Helix-4 I119-Q126 Strand-4 Q129-K135 Strand-5 R137-A150 Helix-5 F156-N160 Strand-6 Strand-1 and Strand-2 Strand-4 and Strand-5 are pretty definite hairpins (and anti-parallel) Strand-1 and Strand-5 have some signal that indicates they could be middle strands. Strand-2 and Strand-4 look like edge strands. Strand-3 has strong signal as edge strand, but some signal as middle. Strand-6 has a very weak signal, as an edge strand -- but most of the CAFASP3 secondary structure predictors do show a piece of strand at the C-terminal end. This is consistent with many arrangments. Ideas: E2 E1 E3 E5 E4 E6 ^ | ^ | ^ | | v | v | v with E2 and E3 connected by H2 E3 and E4 connected by H3 E5 and E6 connected by H4 E3 cannot be an edge strand in this scheme. There could also be two sheets in the structure ? I'll start by putting in N-O constraints for the two hairpins. // hairpin between E1 and E2 constraint I39.N I50.O 2.3 2.7.3.2 constraint I39.O I50.N 2.3 2.7 3.2 constraint L41.N I48.O 2.3 2.7 3.2 constraint L41.O I48.N 2.3 2.7 3.2 //hairpin between E4 and E5 constraint I125.N I130.O 2.3 2.7 3.2 constraint I125.O I130.N 2.3 2.7 3.2 constraint T123.N V132.O 2.3 2.7 3.2 constraint T123.O V132.N 2.3 2.7 3.2 #possible constraints to hold together E3 and E4 E3 and E4 (parallel) E3 KLYSKEIKLKQ E4 KVQTQIQ ?? constraint I96.N Q124.O 2.3 2.7 3.2 constraint I96.O Q126.N 2.3 2.7 3.2 constraint K94.N Q122.O 2.3 2.7 3.2 constraint K94.O Q124.N 2.3 2.7 3.2 constraint Y92.N K120.O 2.3 2.7 3.2 constraint Y92.O Q122.N 2.3 2.7 3.2 This one seems equally good . . . E3 KLYSKEIKLKQ E4 KVQTQIQ For try8, begin with try7-opt.scwrl and add above constraints. 20 August 2002 Rachel Karchin With our latest template library, this target actually has some weakly good hits -- 1dg1g, 1d8tA and 1efuA (which are in the structural family represented by 1efc). These are three domain chains, and the target is similar to the a/b domain c.37.1.8 which has mixed beta-sheet with 6 strands: 231456 with this configuration ^ | | | | | | v v v v v t0148 has six beta sheets, but they are predicted to be anti-parallel 22 August 2002 Rachel Karchin New best scorer is T0148.try8.20.60 Sheet is not assembling yet. E4 and E5 are forming a hairpin, but registers don't look right. E3 is a mess, and oriented away from E4. E1 and E2 are forming hairpin, again the registers look wrong. However E3 is pretty much in one piece, as opposed to try7-opt-scwrl in which it was broken up in two pieces. This would look a lot better if E3 was pulled in and lying next to E4. MIME-Version: 1.0 From: Rachel Karchin Date: Thu, 22 Aug 2002 09:49:09 -0700 To: karplus@soe.ucsc.edu Subject: T0148 Kevin, if you have a chance, take a look at the latest run. It is not looking good yet -- I'm not sure if it has a single sheet, or two (or three?). Latest run has H-bond constraints I put in attempting to form two hairpins (E1 and E2) (E4 and E5) and to line up E3 and E4 in parallel, which is not working yet. -- rk Date: Thu, 22 Aug 2002 10:16:49 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: t0148 is making progress T0148 is making progress. I'd try putting in some CB constraints to get T47-I48 antiparallel to E128-G134. You may want to play with the constraints on S93-Q100, to try to make it be a clean strand. I'm not sure what (if anything) we want to do about F156-F158. Should it be parallel to E128-G134 and antiparallel to T47-I48? If you get a whole new set of constraints, you may (or may not) do better starting over with all the TryAllAlign and no initial conformation. I sometimes will start two job shortly after one another---one with an initial conformation and no TryAllAlign, one with random initial conformation and TryAllAlign. -------------------------------------------------- 22 August 2002 Rachel Karchin I made the turn too short on the E4-E5 hairpin for try8 This will be fixed on try9. try8.20.60 is adding a short strand which is not in our secondary structure predictions (A82-H87). This strand (which I'm calling E2.5) forms a hairpin with E3. For try9, I'm going to assemble a sheet using E2.5, E3, E4 and E5 arranged as ^ ^ | | | | v v with E3 and E4 connected by helix-3. I Q G A E I E E -- K \ Q Q S S / T -- V E -- Y \ Q R H L / V -- V H K \ K G These will be set up as H-bond constraints. 23 August 2002 try9-opt is now the best scoring conformation, but looks worse than try8-opt. The sheet did not form -- and the E4-E5 hairpin is worse -- E5 got pulled apart. I'm going to go back to try8-opt and do a cut-and-paste. A sheet was beginning to form on try8-opt (between E2.5,E3,E4,E5). The E2.5,E3 hairpin just needed to move closer in to the other strands. Perhaps H-bond constraints are a bad choice, since the true topology is unknown. Weaker C-beta constraints may give better results. A83-L98 $x+2,$y-19 On try10, best scoring decoy is still try9-opt.pdb Fri Aug 23 15:00:57 PDT 2002 Kevin Karplus I liked the hairpin in try8-opt which had Q124.O R131.N Q124.N R131.O Q126.N Q129.O I would extend it to include T133 and Q122. The K94, Q124 pairing looks good for strands E2 and E3, but I'd loosen it to have max distance 9 instead of 6, but increase all min distance to 4. In try8, strand 2.5 and strand 4 want to be on the SAME side of strand 3. I like strand 4 much better, so let's lose 2.5. For E1 E2, let's copy the register from try8, hbonding N+O of V38 T51 E40 K49 N42 T47 24 August 2002 Rachel Karchin With latest constraints top three scorers are: T0148.try7-robetta4.19.60.pdb this one has three nice hairpins, but no sheet is forming The third hairpin is the questionable E2.5-E3 pairing T0148.try9-al6.0.60.pdb T0148.try9.14.60.pdb these two don't look as good as robetta.4.19.60 to me and don't have sheet forming either. I'm going to turn down the constraints (0.01) and rescore the decoys to see if anything else pops up as good. With these weights, the top three scorers are: T0148.try9.14.60.pdb T0148.try7-opt-scwrl.pdb T0148.try9-al10.pdb Also try turning up constraints (0.05) and rescoring. (results are same as 0.03 constraint weight) Problem with all of these is that E3 and E4 are not getting lined up. I wonder if we have any decoys where this is the case. I'm going to make a little constraints file of only E3 and E4 and then score the decoys with constraints set really high to see if I can find one like this. T0148.try9-al6.12.60.pdb Doesn't work. The best one has E3 broken up. 26 August 2002 Rachel Karchin T47-I48 antiparallel to E128-G134 ? This would connect E2 and E5 and possibly form a sheet. try12 uses the last three best conformations (above) and current constraints, plus constraints to make E2-E5 anti-parallel ETIKI |||| TVRV // E2 and E5 anti-parallel constraint T47.CB T133.CB 4 5.12 9 constraint I48.CB V132.CB 4 5.12 9 constraint K49.CB R131.CB 4 5.12 9 constraint I50.CB V130.CB 4 5.12 9 try12 hasn't finished -- but the best scoring decoy is try12-try9.27.20 this decoy is not using the constraints I set up to hold E2 and E5 together. They are around 24 A apart (vs. the 9 max). Perhaps constraints need to be weighted more. Can't pick the 5 least trashy models yet, I don't think there's even one that's not trashy at this point. We've got some nice hairpins in the best ones, but no beta-sheet formation. I'm going to start try13 while try12 runs. This try is using the five best scoring alignments, constraint weight is pushed up to 0.2 (from 0.05). It also gets the best conformation so far from try12, T0148.try12-try9.27.20.pdb try13 is now running. Current best is T0148.try13-al4.14.20.pdb E2 and E5 are closer together now, but not close enough to form a sheet, and they don't lie in the same plane (E5 is almost perpendicular to E2) undertaker seems stuck in a local minimum -- the best scoring conformations look very similar and are trashy. The pmodeller model for this target is much better than what we've got. Not great, but it has all the hairpins and a three-stranded sheet. The sheet is antiparallel with these strands. E2.5 E5 E4 To get some variety, I'm going to try a run (try14) which starts from this conformation and does not have constraints. 27 August 2002 Rachel Karchin No luck with try14. Hairpins and sheets exploded. To make this work, probably need to put in constraints to match how the hairpins and sheets line up in the pmodel target. Howver, try13 has produced a promising decoy try13-opt-scwrl in which a beta sheet is starting to form. with no constraints, the scoring function likes try12-try9.8.20 the best other decoys with promise: try13-al1-scwrl.70.20 try13-al2.31.20 From: Rachel Karchin Date: Tue, 27 Aug 2002 09:35:50 -0700 To: karplus@soe.ucsc.edu Subject: T0148 beginnings of a sheet on try13-opt-scwrl. Any ideas how to improve it? I'd like to move the residue 55-70 helix, but am not sure where or how. -- rk Date: Tue, 27 Aug 2002 10:06:55 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu Subject: Re: T0148 It looks like L91-L98 is parallel to V121-I125. Try selecting a good register, maybe gkLYSKEIKLkq skikVQTQIqge The Cterminal strand F156-N160 looks like it should be parallel to G129-T133 and antiparallel to T47-T52, maybe like so: 151> elgqpFQFNNfrd 126> qgeQVRVTGks 55< dseTTIKITenk The D54-K71 helix looks ok to me---it will pack against the face of the sheet if we're lucky. -------------------------------------------------- 27 August 2002 Rachel Karchin 7:48pm best scorer is now T0148.try15-opt.pdb in spite of constraints, cterminal strand has not lined up with its hoped-for partners. N160 and E46 were constrained to be close (4 5.12 9) but are 23.8 angstroms apart in this decoy ???? with constraints weighted very high, best scorer is: T0148.try15.16.20.pdb which is very similar to try15-opt From: Rachel Karchin Date: Tue, 27 Aug 2002 20:03:04 -0700 To: Kevin Karplus Subject: T0148 not progressing My attempts to get a sheet to pack together with constraints are not having much success on T0148. try15-opt is current best. I've got time to do another run tonight, but am uncertain how to get undertaker to move the putative c-terminal strand "E6". current constraints are in try15.constraints. -- rk -------------------------------------------------- The constraints in try15.constraints looked inconsistent to me, and the weakly predicted N-terminal strand seems like it would be useful, but it was omitted in try15.constraints. 1> mpSFDIVse 53< eTIIKITen I've put together a new set of constraints in try16.constraints, guessing the register of this strand. 28 Aug 2002 05:30 Kevin Karplus try16-opt is new best score. Let's try reversing the module 1-51, so that strand E36-L41 is parallel to F156-N160. That will make the long helix have a nice Z-connection across the sheet. and have all the short helices in parallel on the same side of the sheet. 28 Aug 2002 Rachel Karchin try15-opt still looks better than try-16 opt to me, because try16-opt has too many breaks. try17 is still running, best scorer is T0148.try17-try15.5.80.pdb which looks interesting. I like the reversal of 1-51. I don't think we have anything good prior to try15, might want to include try15-opt as last choice, and pick the best variations of try17 to submit. Wed Aug 28 10:21:15 PDT 2002 Kevin Karplus Looking at T0148.try17-try15.5.80.pdb, the helix R16-Y30 looks to me like it is on the wrong side of the sheet. Let's cut off L12-V35 and move it to the other side of the sheet, adding constraints to pair R16-Y30 with helix R137-V147. Try interdigitating V15 V19 N23 S27 V147 L140 I144 Wed Aug 28 13:05:51 PDT 2002 At the group meeting we discussed possible strand orders for the sheet. We guessed directions from whethre there was an intervening helix or not Strand residue direction 0 I6 < 1 E40 |< 2 I50 |> 2.5 I81 >| 3 E95 <| 4 T123 |< 5 R131 |> 6 N159 > There are many possible topologies. One I am going to investigate for try19 is // Strand res direction // 2.5 I81 >| sLDIpaeseh // 3 E95 <| kLKIEKSYLkg // 4 T123 |< gqIQTQVki // 5 R131 |> eQVRVTGks // 0 I6 < sVIDFs // 6 N159 > pFQFNNFrd // 1 E40 |< enLEIVAEvg // 2 I50 |> knETIKITTe Date: Wed, 28 Aug 2002 10:40:26 -0700 (PDT) From: yael@biology.ucsc.edu To: Kevin Karplus Subject: Re: team meeting Wed 11am Sorry, I will not be able to come to the meeting today (have an OB appointment and may have to stay in hospital after) I managed to look at T0147 and T0148 T0149 (mainly at the opt files) Here are my suggestions for models to submitt (ranked by order): T0148 1. try17-try15.5.80.pdb 2. try15-opt 3. try13-opt 4. try8-opt ------------------------------------------------------------ Wed Aug 28 14:56:40 PDT 2002 Kevin Karplus I've noticed that helix G73-S54 needs to turn a quarter turn to bury its hydrophobic face. Maybe I need to add some constraints and breaks? Wed Aug 28 18:41:17 PDT 2002 Kevin Karplus Nope, it's too late and I'm too tired. I changed the score function to get rid of the conjectured packing of the helix (try20.constraints). If I decide to do another run, it will have to be a MUCH shorter one---time is running out. Wed Aug 28 19:23:12 PDT 2002 Rachel Karchin Two topologies are in the running: We are not getting a complete sheet with any of the runs. best scoring from try19 with try20 constraints: T0148.try19-try17-2.7.80.pdb *** strands are lining up nicely but at the price of alot of breaks. two sheets: 0? 1 2 6 u d u u also 4 5 2.5 3 d u u d NOT the desired order strand 0 has not formed well The putative strand 6 at the c-terminus has formed well and is lining up in a sheet with strands 1 and 2. turn connecting Strands 1 and 2 is broken helix connecting Strands 3 and 4 is broken break between R137 and D138 . . . lots more Let's crank break up from 10 to 40 and see what decoys score best T0148.try19-try17-2.7.80.pdb is still the favorite. best scoring with try19 constraints (break back to 10) try19-opt.pdb E1-E2 hairpin not as good as try19-try17 (above) lotsa breaks T0148.try19-try17-2.10.80.pdb T0148.try19-opt-scwrl.pdb -- no reason to prefer either of these with try18 constraints? T0148.try19-try17-2.6.80.pdb T0148.try19-try17-2.3.80.pdb T0148.try18.10.80.pdb T0148.try18-opt.pdb 4 5 1 2 sheet trying to form but failing lots of breaks T0148.try17.2.80.pdb ** similar to best try19, but fewer breaks. No c-terminal strand. T0148.try15-opt.pdb some breaks (not as many as best try19-try17) not as compact hairpins but no sheet T0148.try17-try15.5.80.pdb again some breaks slightly more compact than try15-opt pre-sheet (hairpins only) back to try20-constraints T0148.try19-opt.pdb is best scorer similar to T0148.try19-try17-2.7.80.pdb probably better still lotsa breaks long strands T0148.try19-opt-scwrl.pdb possibly fewer breaks? or are my tired eyes failing? Wed Aug 28 20:35:52 PDT 2002 With no constraints and breaks at 10, try16-opt scores best. It sort of has two sheets, and is more compact. Basically 3 choices 1 T0148.try16-opt.pdb 2 T0148.try19-opt.pdb 3 T0148.try17.2.80.pdb // similar to try19-opt but less broken // and no C-terminal strand