Fri Jun 2 09:27:45 PDT 2006 T0314 Make started Fri Jun 2 09:28:33 PDT 2006 Running on lopez.cse.ucsc.edu Fri Jun 2 13:56:30 PDT 2006 Kevin Karplus No good hits (best with HMM is 1vemA with E-value 5.8) Best with blast is 1peqA (29% id over 70 residues, E-value 0.4). Fri Jun 2 18:16:14 PDT 2006 Kevin Karplus Try1-opt2 looks like a mess. We may have to guess sheets by hand and hand-assemble them. Mon Jun 19 12:34:19 PDT 2006 Grant Thiltgen I started try2. I edited out the automatic constraints. I put in the two big helix constraints and I put in the two strands that are the most predicted. str2 has them in anti-parallel, so I put the sheet constraint in for antiparallel. Since they're separated by a helix, I am going to try a run that has them in parallel as well. try3 will have them as parallel. Mon Jun 19 17:57:41 PDT 2006 Grant Thiltgen try2 looks alright. try3 looks pretty crappy but scores pretty well. I'm running try4 based on try2.costfcn and trying to get the sheets that I want. I'm also running try5 to try to get the correct buried resides close to each other. Tue Jun 20 13:30:52 PDT 2006 Grant Thiltgen try5 looks okay. I like try2 and try5 the best. I mostly get different structures every time. I'm not sure how I'll be able to build this protein. I looked in the PDB for lipoproteins, but it gives me mostly proteins related to lipoproteins. I found a lipoprotein in E. coli that is of similar length, so I may try modelling the protein after that one. The PDB name is 1oap. I'm running a VAST search to see if there are others, and I will then run try6 with the hits as a top alignment to try to get the protein to align to the structure. I got two more structurally similar proteins from VAST. Tue Jun 20 16:26:09 PDT 2006 Grant Thiltgen try6 doesn't look too bad, but it's not exactly the same shape as the protein that I tried to start with. I'm not exactly sure how to make it look like that exactly. I'll do try7 with the same costfcn to see if I get different results. Wed Jun 21 13:24:45 PDT 2006 Grant Thiltgen I REALLY like try7. It's pretty close to the structure for the E. coli lipoprotein of similar length. I'm going to try tweaking it a bit, keeping the sheet that is there and seeing if I can manage to get a fourth strand to form in the sheet. I'm running try 8 with the parts I liked and see what happens with the rest. If it doesn't work I can try to force a small strand at the beginning. Wed Jun 21 20:45:23 PDT 2006 Kevin Karplus There needs to be a bit more attention paid to burial on this one---I8 is badly exposed, but could be tucked down near V57 without too much difficulty. Please make up a list of models for me to submit (soft deadline) by tomorrow (Thursday) afternoon. Date: Thu, 22 Jun 2006 14:15:21 -0700 From: Kevin Karplus To: thiltgen, martin.madera CC: karplus Subject: T0314/README I was trying to put some comments in the T0314.method file for the soft submission, and found almost no notes in the README file. Please, make notes of what *EVERY* try is attempting to do, and what the results of it are! Also, Grant asked for try9-opt2 in the superimpse-best.under, but there is no such file. Grant and Martin, please try again on the T0314 list of top models, including only models that exist. Also, please update the README file---it is our lab notebook and like all lab notebooks should be maintained as the work is done, not after the fact. Kevin Thu Jun 22 15:48:39 PDT 2006 Grant Thiltgen I swear I wrote in here about try9, but I must have no saved or something. For try9, I tried to make a small strand in the sheet like the protein I'm modelling it after has (1oap). I am starting try10 to try to bring the n-terminal end into the strand and bring the protein together on that end, since that end sticks out a bit on try8. Thu Jun 22 16:25:30 PDT 2006 Kevin Karplus There does not seem to have been any analysis of the server predictions done for T0314. The top-scoring ones with unconstrained costfcn are Pcons6_TS1, ROBETTA_TS1, Pmodeller6_TS2, ... These look as good as the ones we've been builing, at least at first glance, so we might want to give them a chance also. I'll submit the list Grant gave me, but models 1, 2, and 5 are too similar when we have so little readon to believe we know what fold to submit. Mon Jun 26 14:01:36 PDT 2006 Grant Thiltgen I started try11. Try11 is attempting to rebuild the whole protein from scratch using the three templates from E. coli. I increased the phobic fit and dry weights in order to attempt to bury the exposed residues better than they were in the previous attempts. I am also looking at the top scoring server predictions. I'll work on using them in order to create some new trys in order to get some diversity with our predictions. Wed Jun 28 17:21:21 PDT 2006 Grant Thiltgen I am starting try12 on camano. I put the alignments from all the predictions back in as well as the three new proteins to see if I can get anything new and different for submission by the hard deadline. Sat Jul 1 19:28:17 PDT 2006 Grant Thiltgen Try13 is a polishing run on the Robetta models (TS1-TS5). Try14 is doing one on the other top scoring server model, Pcons6_TS4. Both trys are started on whidbey. Sun Jul 2 13:09:32 PDT 2006 Grant Thiltgen Polishing the Robetta models scored better than try8, which is our current best model. Try14 which is the Pcons repack scored slightly worse than try8. I am going to run two more undertaker runs to see what other interesting configurations I can get and try polishing some of our models to see what that difference is. Try15 and try16 are started on whidbey. I included the original alignments and only the lipoprotein from my pdb search. I also used the original dssp-ehl2 constraints from the first run. Sun Jul 2 21:38:18 PDT 2006 Grant Thiltgen Try15 and try16 look mediocre. I'm not sure either of them are all that great, but try15 scores better than try16. I think I may have to pick a few, and the two polished models from the two best scoring server models and polish them. Mon Jul 3 12:29:53 PDT 2006 Grant Thiltgen Currently, I have decided to polish models try13, try8, try14, try6, and try16. They are all different models. Right now, try17 is a polish of try8, try18 is a polish of try6, and try19 is a polish of try16. try17 and try18 are started on camano. try19 is started on whidbey. Sun Jul 9 20:47:59 PDT 2006 Grant Thiltgen Currently, for the best models, I have try17, try13, try14, try18, and try19. They are all fairly different. Make started Mon Jul 10 16:11:05 PDT 2006 Running on vashon.cse.ucsc.edu I ran this to get the sep predictions for this protein. Mon Jul 10 17:09:16 PDT 2006 Grant Thiltgen After getting the sep predictions, I decided to start from the beginning and attempt to build the sheets and helices predicted from the str2, n_notor, o_notor, o_sep, and n_sep predictions. Try20 is starting over from the beginning using new sheet and helix constraints. Try20 started on orcas. Tue Jul 11 11:49:37 PDT 2006 Grant Thiltgen Try20 looks really good actually, except for one strand that is hanging loose. I looked at things in burial, and it appears that there's a gaping hole. I think if I move the free strand up to pair with the anti-parallel sheet, that would fix the problem. I'm working on making the sheet complete in try21. Tue Jul 11 12:17:52 PDT 2006 Grant Thiltgen Try21 is ready. I moved the last strand up to try to complete the sheet from try20. I also decided to try to get the last strand from the str2 predictions to form. I am also going to run try22, which will not make the last strand form from the str2 alphabet, but instead just use the helix going across the sheet. Try21 started on camano. Tue Jul 11 12:24:55 PDT 2006 Grant Thiltgen try22 is ready, it has one last strand on the sheet and a longer helix. I'm running try22 on camano. Tue Jul 11 13:21:53 PDT 2006 Grant Thiltgen Wow. try21 and try22 are looking pretty bad. I'm not sure what I did, but the sheets are all over the place. I'm going to start try23 and try24. Try23 is the same constraints as try21, but with the constraints in the costfcn turned up. Try24 is the same as try22, with the constraints in the costfcn turned up. Try23 and try24 started on whidbey. Tue Jul 11 13:30:18 PDT 2006 Grant Thiltgen I see what happened, I think I had some conflicting constraints. I forgot to edit out one of the sheets from the try20-opt2.sheets file. For try25, I dropped the main constraints back to 10, and I increased the sheet weights and the helix weights that I wanted to keep. I added the new sheet constraints and gave them a higher weight. I included the helix that cross sideways across the sheets, but I kept a lower weight on that. Try25 started on lopez. Tue Jul 11 14:37:08 PDT 2006 Grant Thiltgen Hmm. I still can't get that sheet to form. I'm going to work on try26 which is only going to add one strand to the sheet. I am going to try to extend the current fourth strand out by one residue in order to make it line up with the current third sheet. So for try26, I made the last strand on the sheet longer, and attempted to pair it with the strand that's pairing by the helix. Try26 started on camano. Tue Jul 11 15:08:05 PDT 2006 Grant Thiltgen So far the top models I have are try17, try13, try14, try18, and try20. I'm still working on getting try20 fixed to get the sheet included. I'm now going to work on modifying try13 to include the last strand in the sheet. try27 started on whidbey. Tue Jul 11 15:35:10 PDT 2006 Grant Thiltgen Gah! still no sheet. I think I'm going to remove all the helix constraints just to see if I can get the sheet to form without any helices. Who knows what it is going to do though. Maybe I'll try starting from the model and keeping the the things I want and getting the new stuff involved instead of starting from scratch. Try28 just has the sheet constraints and is starting from the alignments. Try28 started on lopez. Tue Jul 11 15:41:15 PDT 2006 Grant Thiltgen Try29 is going to bring in the try20-opt2 model and try to form the sheet I want from that. try29 started on heehaw. Tue Jul 11 16:03:35 PDT 2006 Grant Thiltgen try27 didn't look anything like try13, so I'm running try30 which has the same costfcn, but using try13-opt2 as a model to start from. try30 started on whidbey. Tue Jul 11 18:08:28 PDT 2006 Grant Thiltgen Try30 doesn't look much different than try13 which it is modelled after. Perhaps the strand is closer to making a sheet than it was before. Try31 is going to take try30 and put a higher weight on the SheetConstraint to make the sheet form. Try31 started on camano. Tue Jul 11 18:14:36 PDT 2006 Grant Thiltgen Try29 did bring the strand up closer to the sheet, but it still hasn't formed the sheet. Like I did in try31 for the other model, I am going to increase the strand constraint on the strand and use the try29 model in order to try to get the strand to form. Try32 started on camano. Wed Jul 12 09:55:58 PDT 2006 Grant Thiltgen Try31 seemed to have messed up the protein a bit. Try32 didn't really do much to fix the sheet. I'm going to look into manually moving the sheets using protein shop. Some of the runs I've been making have been scoring well on the unconstrained costfcn, even though they still have some problems, so I'm hoping that once I fix the problems and look more protein-like it will score well also. Wed Jul 12 10:11:08 PDT 2006 Grant Thiltgen I moved the strand in try20-opt2 to form with the sheet in proteinshop. I moved the helix that runs perpendicular across the sheets to be a bit closer to the sheets, since proteinshop moved it a bit out of the way. There are some clashes, but I've been lucky with undertaker keeping things together, so I'm going to attempt to run it through undertaker and clean things up a bit. Try33 is taking try20-opt2-moved-strand-and-helix.pdb and trying to optimize it. I am trying to keep the sheet together while adding the fifth strand. Try33 started on camano. Wed Jul 12 10:33:03 PDT 2006 Grant Thiltgen Try34 is taking try13-opt2-moved-strand-and-helices.pdb and trying to optimize it. It still has a lot of clashes, so I'm hoping it's not too much for undertaker. I'm hoping to add the fourth strand to the sheet in this model. Try34 is started on shaw. Wed Jul 12 10:53:02 PDT 2006 Grant Thiltgen Currently, my top5 are try21, try17, try13, try14, and try20. I'm hoping if I get the "fixed" models of try17 and try20 I can push them further up the list. Try21 was actually a "mistake" from a run trying to get try20 fixed, and it looks pretty decent. I was a little concerned about the structure, but a VAST search on the PDB file showed me that it is a structue that occurs in parts of proteins, so it's not as far off as I thought. I'm going to work on polishing try21, try17, and try14 while I'm working on completing the other two models. Wed Jul 12 11:05:07 PDT 2006 Grant Thiltgen Try35 is a polish of try17. Try36 is a polish of try14. For these I upped the weights of soft_clashes to 60, breaks to 200 and dry6.5 to 40 and dry8 to 35. Try37 is a polish of try21. For this I upped the weights of soft_clash to 40, break to 100, and dry6.5 to 30 and dry8 to 25. Try35 started on shaw. Try36 started on orcas. Try37 started on camano. Wed Jul 12 11:52:04 PDT 2006 Grant Thiltgen I'm going to attempt to manually modify the loop region in the T0314.try20-opt2-moved-strand-and-helix.pdb file to fix the really bad loop region that protein shop made. If it works well I will also try to manually fix the loop in the undertaker model that I polished from that proteinshop model. Wed Jul 12 12:01:30 PDT 2006 Grant Thiltgen That worked really well. Now I have a big gap between the loop and the strand that I want as part of the sheet, but it should work in undertaker (I hope!). I am going to raise the reduceconstraint and optsegment arguements in the first optconform in order to get the constraints right, then fix the loop. This will be try38. Try38 is ready and started on shaw. Wed Jul 12 12:23:19 PDT 2006 Grant Thiltgen I also decided to break the loop in a different location in order to try to get the strands to form. Try39 will use the same costfcn as try38, but with the different break in order to try to get the sheet to form. Try39 started on camano. Wed Jul 12 13:36:25 PDT 2006 Grant Thiltgen I'm not sure if try38 or try39 is going to help, but if it gets something there maybe I can close the gaps afterwards. I forgot to remove constraints from the costfcn for polishing, so I think my polish runs didn't work, so I need to restart those. Try40 is a polish of try17. Try41 is a polish of try14. For these I upped the weights of soft_clashes to 60, breaks to 200 and dry6.5 to 40 and dry8 to 35. Try42 is a polish of try21. For this I upped the weights of soft_clash to 40, break to 100, and dry6.5 to 30 and dry8 to 25. Try40 and try41 started on shaw. Try42 started on lopez. Wed Jul 12 14:01:49 PDT 2006 Grant Thiltgen Try33 is very close to what I wanted try 20 to look like. This is a possibility, but I am going to work on getting the strand to form correctly. I am going to work on it by itself, and with replacing the loop region. Try43 is working on modelling try33 without forming a break in the turn, which may cause a problem, but I think it's worth trying. Try43 started on camano. For the replacement of the loop region, I think I am going to use the intial model, try20-opt2, and try33-opt2, using up to residue 45 of try20 and the rest from try33. Try44 started on camano. Wed Jul 12 14:20:46 PDT 2006 Grant Thiltgen Try34 ended up making a helix where I wanted a strand. I am going to run try34 again with a higher sheet constraint. Try45 is the rerun of try34 with a higher sheet constraint. Try45 started on vashon. Wed Jul 12 16:21:49 PDT 2006 Grant Thiltgen Yes! Try44 finally started forming the sheet. I think if I add a break in the other half where the helix got torn apart, I may be able to get the whole sheet to form. The hairpin may be a bit off, and I need to check to make sure that I didn't screw up the residues there, and that my strand is off by one. That would be annoying if I did. Nope, the strand is okay. I am going to try to break it one more time between residue 46 and 47 in order to get the beginning of the sheet to form. I am also going to break the coil at the end of the strand to get the other end of the strand to form. I am going to cut the models between residues 57 and 58. I may have to superimpose three different models to get that to happen. Or at least copy and paste different sections the two models together. Wed Jul 12 16:41:39 PDT 2006 Grant Thiltgen I decided to use residues 47-57 of the new model and the rest of the residues from try20-opt2 to create my model. I'm hoping that it will clear things up and make the entire strand in the sheet I'm looking for. This is try46 and it is running on shaw. Wed Jul 12 16:45:21 PDT 2006 Grant Thiltgen My current top models are try44, try30, try21, try35, and try14. I am hoping to have slightly better models of try44 and try30, but I have no guarantees. I think for fixing try30, I am going to make the strand constraint stronger and lengthen the constraint one residue on each end of the strand. I am also going to introduce some breaks in order to get the strand to orient correctly. Wed Jul 12 17:13:45 PDT 2006 Grant Thiltgen I edited the try30-opt2 model to have the strand that I moved in proteinshop. There are breaks on either side and I am planning on having undertaker try to form the sheet. This is try47 and it is started on shaw. Wed Jul 12 17:40:09 PDT 2006 Grant Thiltgen Try46 formed the helix again in the loop region that I didn't want. I think I am going to have to increase the sheet constraint even more to get it to form. In try48 I have increased the sheet constraints on the sheet I want to form, and I'm using the same model with the two breaks that I tried optimizing from try46. Try48 started on whidbey. Wed Jul 12 17:59:23 PDT 2006 Grant Thiltgen Currently, these are the top five models: T0314.try44-opt2.pdb T0314.try30-opt2.pdb T0314.try21-opt2.pdb T0314.try35-opt2.pdb T0314.try14-opt2.pdb Here's a history of these models: T0314.try44-opt2.pdb is the closest match to the secondary structure predictions. I made a hairpin of residues 20-29 because of strand predictions and separation predictions that looked good from the sep alphabets. The helix constraints are straight out of the helix results. I made the sheet constraints from the str2 secondary structure predictions. I made a small sheet constraint that I ended up removing later. The first run of these constraints (try20.costfcn) gave me a very good base to start from, but there was an extra strand that wasn't pairing up with the rest of the sheet. I finally ended up moving the strand manually with ProteinShop. I ended up having to add breaks to the models in order to get the strands to form correctly. Try30-opt2 came from try13-opt2. Try13 was a polishing run on the five Robetta server models. I ended up getting something vaguely proteinlike. Try21-opt2 was an attempt to get the strand in place from try20-opt2, but it didn't work well. It ended up looking decent on its own, so it's a valid model. Try35-opt2 is a polished model based on alignments from a lipoprotein I found in the PDB. A lipoprotein in E.coli (1oapA) was of similar length and I got a few other structures from vast (2aizP,1r1mA) to make alignments from, and I got a structure that looked somewhat like the E.coli lipoprotein. Try14-opt2 was a polish of the best scoring Pcons6 model (TS4) in undertaker. I ran a polishing run on it and this is the model. Wed Jul 12 19:37:16 PDT 2006 Grant Thiltgen Try48 looks the best of the models that I have managed to get for the sheets so far. Try47 didn't work so good, but I will increase the constraints and try to run try49 on that. Try49 is going to increase the constraints of try47 to try to get the sheet to form, but it may not work, and we may just have to go with try30. Try50 is going to be a "polishing" run on try48. I want to try to bring the helices in closer. Try49 and try50 started on shaw. So currently the top models are: T0314.try48-opt2.pdb T0314.try30-opt2.pdb T0314.try21-opt2.pdb T0314.try35-opt2.pdb T0314.try14-opt2.pdb Here's a history of these models: T0314.try48-opt2.pdb is the closest match to the secondary structure predictions. I made a hairpin of residues 20-29 because of strand predictions and separation predictions that looked good from the sep alphabets. The helix constraints are straight out of the helix results. I made the sheet constraints from the str2 secondary structure predictions. I made a small sheet constraint that I ended up removing later. The first run of these constraints (try20.costfcn) gave me a very good base to start from, but there was an extra strand that wasn't pairing up with the rest of the sheet. I finally ended up moving the strand manually with ProteinShop. I ended up having to add breaks to the models in order to get the strands to form correctly. Try30-opt2 came from try13-opt2. Try13 was a polishing run on the five Robetta server models. I ended up getting something vaguely proteinlike. Try21-opt2 was an attempt to get the strand in place from try20-opt2, but it didn't work well. It ended up looking decent on its own, so it's a valid model. Try35-opt2 is a polished model based on alignments from a lipoprotein I found in the PDB. A lipoprotein in E.coli (1oapA) was of similar length and I got a few other structures from vast (2aizP,1r1mA) to make alignments from, and I got a structure that looked somewhat like the E.coli lipoprotein. Try14-opt2 was a polish of the best scoring Pcons6 model (TS4) in undertaker. I ran a polishing run on it and this is the model. Wed Jul 12 20:37:05 PDT 2006 Grant Thiltgen I don't think try49 or try50 are going to work. I think the list we have is going to have to be it. Wed Jul 12 20:55:46 PDT 2006 Kevin Karplus I submitted Grant's top hits: ReadConformPDB T0314.try48-opt2.pdb ReadConformPDB T0314.try30-opt2.pdb ReadConformPDB T0314.try21-opt2.pdb ReadConformPDB T0314.try35-opt2.pdb ReadConformPDB T0314.try14-opt2.pdb At least they are a diverse group of models! Sun Sep 24 18:23:46 PDT 2006 Kevin Karplus As of today, this looks like one of the hardest targets, with nobody getting GDT over about 19%. Thu Mar 22 10:17:07 PDT 2007 Kevin Karplus With GDT being computed better, the top GDTs are around 31.55% (Zhang-Server_TS5). Our best model was try31-opt2.repack-nonPC (not submitted). Our best submitted was model2=try30-opt2, which was slightly worse than what our server submitted, but still not bad compared to other servers (it was a polishing of try13, in turn from robetta). The SAM_T06_server was second of the TS1 models (after PROTINFO-AB).