Kevin Karplus 9 May 1998 The target98 method found NO homologs for t52 (other than itself), so the target model will not be very good for this target. We may want to use double-blast as well, since it is not likely to be any worse than target98 with a single sequence. We do get a fairly good score for sequence 1pmd (peptidoglycan synthesis multifunctional enzyme), though not to its structural homologs 1btl, 2bltA, and 3pte. Of course, 1pmd is large, and t52 is small, so the homologs may be to different domains. Searching in the other direction, using template models, I get no strong scores. The strongest (1hsq) is only -2.460. Summing in both directions gives as the top 4: 1pmd -6.210 1hsq -4.51 1pdgA -3.25 (1pdgB, 1pdgC) 1broA -2.830 (1broB) -6.2 is in the range where 15 more true positives gives 40 more false positives in the chothia/domains test, and 21-33 more true positives gives 50 more false positives in the fssp test. So the probability to assign to "new fold" is about 60-75% (probably around 70%). The alignments of 1pmd to t52 using the t52 model then retraining look pretty good (as alignments---I haven't looked at them in 3-d yet). If they look good in 3-space, I might be willing to take 1pmd as a prediction, reducing the bet on "new fold". I haven't tried building the 1hsq joint models yet, because, even though this scores in the top few hits in both directions, the score in both directions is VERY poor. -------------------------------------------------- Kevin Karplus 11 May 1998 The t52-1pmd and t52-1pmd-global alignments are only moderately compact, but crosses a domain boundary. I'd like to try with 1pmd fssp alignments, to see if they align any better. 1pmd-t52 is a tiny fragment and 1pmd-t52-global is a scattered set of fragments, though the biggest one is consistent with the t52-pmd alignments. The 1hsq alignments seem to match only a tiny fragment: WQPSNFIE WFPSNYVE A rather fancy strand-1turn helix-strand motif. There doesn't seem to be a corresponding structure in 1pmd. This one is going to be a tough call. From karplus@cse.ucsc.edu Mon May 11 17:59:14 1998 Return-Path: karplus@cse.ucsc.edu Date: Mon, 11 May 1998 17:59:13 -0700 From: Kevin Karplus To: markd@cse.ucsc.edu Cc: karplus@cse.ucsc.edu Subject: another alignment Could you make alignments for 1pmd and put them in ~/pcem/pdb/1p/1pmd/struct-align? (Once you get the Makefile.models97 updated, I'll stop asking you to make alignments, since I can then make them myself.) Kevin ------------------------------------------------------------ 24 May 1998 Kevin Karplus 1pmd-t52-const-global and 1pmd-t52-fssp-global look ok for the main alignment, but the two fragments near the end need to be swept into the middle. The result is not a gret alignment bu may be ok. The unsupported helix at the beginning should probalby be chopped off. 1hsq has an interesting motif that matches very well, but it is quite short, and so is probably not suitable as a fold-recognition target. If we were doing homology modeling, we might try doing some cut-and-paste of 1pmd and 1hsq. 1pmd score gives chothia domains 60 T 200 F 75% new-fold fssp 54.35 T 100 F 65% new-fold From compbio.casp-request Wed Jun 3 17:19:25 1998 Return-Path: karplus@cse.ucsc.edu Date: Wed, 3 Jun 1998 17:19:23 -0700 From: Kevin Karplus To: compbio.casp@cse.ucsc.edu Subject: t52 Although target t52 has been moved up to Jun 10, it looks like we're going to be putting in a "new-fold" prediction. Our best scoring prediction had a rather non-compact alignment, and the two known cystine bridges were not close in the template. Our second-best was a very short piece, with maybe 10 residues well-aligned. The unusual piece of super-secondary structure that these residues represent may or may not be worth reporting. 4 June 1998 Kevin Karplus I'm worried because the target98 alignments did not pick up the swissprot sequence CVN_NOSEL, which is the best hit with BLAST. Maybe the target98 alignments were done on a day when NRP was not correct??? (The log file says Apr 24). I'm going to move them to a "maybe-bad" subdirectory and remake them. Hmm---repeating the search does not find CVN_NOSEL---is it still missing from NRP? Hmm--egrep didn't find it either. I suspect that NRP isn't being updated as quickly as we might like. Maybe I should recheck after Sunday's download.