Thu Jul 11 12:49:35 PDT 2002 11 July 2002 Kevin Karplus T0151 is clearly b.40.4.3, with E-values < 1e-38. The best template appears to be 1qvcA. I'm afraid that undertaker will make a mess of the C-terminal region, since that is shown as being "out in space" in 1qvc. The C-terminal region is not well conserved though, so it may fold up quite differently. The N-terminal region forms a pretty tetramer in 1qvc, with the first strands of adjacent molecules being antiparallel (making the 3-srand antiparallel sheet into a 6-strand one). This is one target where we may be better off not letting undertaker mess with the structure, since it does not know about multimers, and will probably fail to keep the edge of the first strand free to dimerize. In try1-opt, the second strand has gotten detached---we may need to add some constraints to keep it in place. The last strand has also been peeled off and coiled up. We may need to add constraints there also. The first edge is hidden a bit by the loop from T36 to A52, as feared. 12 July 2002 Kevin Karplus With the new fragments, try2-opt looks better. Only the first strand L19-T22 has been separated a bit from where it belongs. Perhaps a few constraints are needed. 23 July 2002 Kevin Karplus Changed the score function to use the new pred_alpha2 cost function. With this function, try2-opt still comes out on top. The scoring function is crumpling up the two hairpins that extend out of the barrel, so we may have to go with an alignment on this one. Looking at the alignments, I'm not real impressed with models 2 and 22 (both from 1qvcA). The simple alignment (model 38 from T0151-1qvcA-simpleSW-adpstyle5.pw.a2m.gz) seems to be one of the better ones. try3-opt-scwrl scores well, but crumples up the tail and ruins one of the hairpins---I think this will HAVE to be an alignment-only prediction, unless we miraculously get a way to handle multimers. 24 July 2002 Kevin Karplus The alignments look pretty good up to about D104, then there is junk that may need to be modeled as a separate domain. I'll create two subdirectories: 1-109 (the strand is predicted to end at S109) and 110-164. I'm not sure that the 110-164 domain has a good fold---in 1qvcA it meanders somewhat randomly over to chain C. We may have to be content with predicting the first domain. 24 July 2002 Kevin Karplus Except for having folded over the second hairpin, the 1-109/ try1-opt-scwrl looks pretty good. The 110-164 try1-opt looks like trash, as expected. I think we'll have to omit this region, since it is not highly constrained and is unlikely to be conserved from the homologs. To do: look at the different models for the 1-109 region, and try to pick the best one. Maybe add constraints to try to keep the hairpins from folding over? 27 July 2002 Kevin Karplus Let's pick some constraints from the alignments: hairpin1 base: A29-I59 hbond hairpin1 middle: S35-L53 hbond hairpin1 end: P37(O)-G50E51(N), I39(O)-K48(N) tip approx T44 constrain tip1 to approx same distance from base1 ATOM 313 CA THR A 44 16.004 11.400 44.534 1.00 0.00 ATOM 191 CA ALA A 29 12.666 53.554 42.957 1.00 0.00 42.315 Angstrom hairpin2 end: F88-K95 hbond hairpin2 base: R82-V101 hbond tip2 approx E92 constrain tip2 to approx same distance from base2 ATOM 607 CA ARG A 82 -0.357 47.093 38.150 1.00 0.00 ATOM 699 CA GLU A 92 19.746 47.186 19.764 1.00 0.00 27.243 Angstrom constrain tip1-tip2 ATOM 313 CA THR A 44 16.004 11.400 44.534 1.00 0.00 ATOM 699 CA GLU A 92 19.746 47.186 19.764 1.00 0.00 43.683 Angstrom short hairpin end: R20-V28 hbond short tip approx S24 constrain short tip-tip2 ATOM 699 CA GLU A 92 19.746 47.186 19.764 1.00 0.00 ATOM 164 CA SER A 24 10.654 65.894 35.274 1.00 0.00 25.946 Angstrom Note: G50 and E51 are superimposed in the alignment, since G50 is inserted. This insertion should fix some of the bulging in hairpin1, pairing P37-E51 and I39-D49. Under 1-109/ the try2-opt-scwrl run gets good, scores, but it indicates that I got the hairpin wrong for hairpin2---I'll try again with S80-D104, R86-R96, T90-E94. Perhaps these come from the 1eygA alignment? 1-109/ Try3-opt-scwrl looks pretty good, though the hairpin centered around 44 is still pretty raggedy. Also, I was wondering if I99 should be close to L55 or R56---it is pretty far away. In the alignment with 1qvcA, D100 is hbonded to O of R56 and N of N58. Let's add some constraints to keep the barrel closed, and try adding some to remove the bulge in the first long hairpin. Increasing the weight of the constraints makes try3-al4.18.30 in 1-109/ score best. Let's start from several of the low-cost conformations and re-optimize. 1-109/ try4-opt-scwrl is the new lowest cost. It's not too bad, but the insertion at G50 E51 A52 is very badly done. Perhaps we need to loosen the constraints on the location of T44. It would also be good to add some more of the h-bonds in the barrel as constraints, as some of them have gotten broken. 29 July 2002 Kevin Karplus For T0151, I think we'll just predict for residues 1-109, since the remaining residues are very glycine-rich and form a rather irregular loose random-looking arrangement in 1qvcA. Because 1qvc is a tetramer, and undertaker can't handle multimers yet, I've been trying to add constraints to hold the dimerization interface fixed. The AB interface consists of the first strand (antiparallel V5-V11 in 1qvcA) and the two long hairpins, which don't touch, but have to stay wide enough apart to let the other half of the dimer in. The AB-CD interface is a large flat beta sheet formed by the first, middle, and last strands. I think I need to add some more constraints to keep the strands from peeling off the barrel. 29 July 2002 Kevin Karplus I'm not sure that 109 is the best stopping place if we are going to do just the barrel domain. In some of the alignments it looks like 111 would have made a cleaner end to the last strand. Perhaps I should start over with 1-115 as the domain to predict, so that 2ry predictions can be reasonable around 109-111. In setting up the constraints for 1-115/define-score.script, I noticed some errors in the atom numbering for earlier constraints. I think I'm going to have to rewrite the CONSTRAINT command to take specifications of the form R56.O instead of "424", since entering the constraints seems very buggy. Also, in 1-115, I'll include a few other templates for b.40.4.3 structures: 1otcA and 1quqA, which have bery different multimerization. The 1otcA structure includes bound nucleic acid and has 3 copies of domain linked into one protein. In looking at the constraints, I'm really quite confused about the pairing for hairpin2, which seems to be different based on different alignments. I may have to make those constraints weak (and even conflicting), until I start getting a good structure. Tue Jul 30 11:07:50 PDT 2002 Kevin Karplus 1-115/decoys/T0151.try1-opt-scwrl looks pretty good. It allows me to diambiguate some of the constraints on the hairpins. Putting in some of the hbonds I see naturally leaves try1-opt as the lowest cost. The GP at 107-109 naturally breaks the beta sheet (the proline is what would be the hydrogen bonding position, which it can't, of course, do). Perhaps we should report just through G107? I tried superimposing try1-opt with the alignment from 1qvcA, and it seems to be good out to about A113, so I won't bother trimming. The ends of the hairpins (at Q43 and R91) have moved a bit, but not outrageously. It may be better to have left them alone, but I'm not unhappy with the small amount of movement. I'm ready for try2. Tue Jul 30 15:19:25 PDT 2002 Kevin Karplus The 1-115/ try2 iterations are progressing well. It looks like I may want to add a few more scaffolding constraints on the ends of the hairpins---perhaps W47 to E100 and E92 to L55---to triangulate a bit more and keep the ends from flopping over. Tue Jul 30 16:51:10 PDT 2002 The 1-115/ try2-opt structure looks pretty good. There are a couple of gaps that need closing, and I'll add the anti-flop constraints before doing try3. From T0151-1qvcA: CA TRP A 47 11.614 18.272 47.291 1.00 0.00 CA GLU A 100 5.054 45.634 38.831 1.00 0.00 29.382 Ang CA LEU A 55 9.580 40.463 40.594 1.00 0.00 CA GLU A 92 19.746 47.186 19.764 1.00 0.00 24.134 Ang From T0151-3ullA: CA LEU A 55 9.868 40.717 40.571 1.00 0.00 CA GLU A 92 11.929 47.500 14.153 1.00 0.00 27.353 Ang I added these constraints, plus dry5, and increased the break penalty: Constraint 340 771 24 29.382 34 // CA W47 E100 Constraint 408 707 21 25 30 // CA L55 E92 1-115/ try2-opt still scores best, so let's start try3. Wed Jul 31 08:49:53 PDT 2002 Kevin Karplus 1-115/ try3-opt improves the score a bit, mainly from running a lot of "OptSubtree". There needs to be some CloseGap run to close the P23-S24 gap. There are some interesting charged surfaces appearing on both the barrel and the arms that may help explain the dimerization. I suspect that solving the dimer would require a score function that includes salt bridges. I'll do one more run, this time turning up break costs and increasing the probability for CloseGap. Changing the score function does move a different decoy to the lowest-cost position (1-115/ try3.11.50.pdb), but the differences are fairly small. I think I'll probably go with whatever try4 comes up with, after trimming any residues off the C terminus that look bad. Wed Jul 31 13:37:59 PDT 2002 Kevin Karplus 1-115/ try4-opt is NOT the lowest scoring when output and read in again, try4.6.40.pdb is. The difference in scoring is probably because of different reconstruction of the breaks at 45 and 48. Maybe I should add a couple more scaffolding constraints, for residues a little farther from the ends of the hairpins, and soften the scaffolding, so that the breaks can come closer to closing. R86-L53 and E18-I39 CA GLU A 18 17.879 51.369 44.346 CA ILE A 39 5.891 20.722 46.358 32.970 Ang CA LEU A 53 10.197 34.161 41.356 CA ARG A 86 8.029 48.674 28.914 19.239 Ang Adding these constraints and widening the range for the other scaffolding constraints still leaves try4.6.40.pdb as the best. Lowering the weights for all constraints and increasing the penalty for breaks doesn't change it either. Let's reoptimize from the top few decoys. Thu Aug 1 10:11:20 PDT 2002 In 1-115/ try5-opt only reduces score slightly relative to try4.6 The helix near A65-S70 is not packed tightly against the end of the barrel. In fact the whole barrel has expanded slightly and has a hole down the center, a problem shared by try4.6 also. Let's do one more run, increasing the weight for dry5, dry6.5, and dry8, to try to get tighter packing. Rescoring with the changed weights moves try5.17.50.pdb to the top Let's reoptimize with a few of the best decoys. 1 Aug 2002 Kevin Karplus Aside from a slight hollowness, 1-115/ try6-opt is looking pretty good. I think we'll submit this one. There are 9 breaks (all small).