Mon Jun 12 10:20:32 PDT 2000 Kevin Karplus T0099 Blast finds strong homology to 1shf[AB], 1fynA, 1ny[fg], 1a0nB, 1azgB, 1avzC FSSP reps are 1bbzA (except for 1shfB, which is 1ckaA) Double blast gets same set of best hits. T0099 looks like a deliberately mutated FYN proto-oncogene. The question is whether the mutations will change the structure. The t2k alignment contains lots of PDB structures, so we should probably back off to a more restrictive set of sequences---either the t2k_1 set, or an iterated alignment with very strict thresholds. Top hits with template models are chainID E-value fssp rep 1pht 6.2e-12 1ckaA 1awj 1.8e-11 1awj 1aojA 1.2e-10 1aojA 1hsq 1.6e-10 1ckaA 1shfA 3.9e-10 1bbzA 1ark 4.1e-10 1bb9 1cskA 4.2e-10 1ckaA 1bbzA 5.8e-10 1bbzA 1awo 8.2e-10 1bbzA 1bb9 9.6e-10 1bb9 1aboA 1.0e-09 1bbzA 1qlyA 1.6e-09 1qlyA 1ckaA 2.4e-09 1ckaA 1shg 2.9e-09 1ckaA 1semA 3.1e-09 1ckaA 1a81A 2.5e-08 1a81A 1griA 2.9e-08 1griA 1tuc 2.7e-07 1tuc 2shpA 2.3e-06 1cviA 1tud 2.9e-06 1tud FSSP equivalence classes: 1ckaA, 1bbzA, 1bb9, 1griA 1awj 1aojA (Z=2.6 to 1ckaA) 1qlyA 1tuc (Z=4.1 to 1ckaA, 4.6 to 1griA, 4.0 to 1bbzA, 2.7 to 1bb9) 1tud (Z=5.4 to 1ckaA, 5.3 to 1griA, 4.6 to 1bbzA, 2.7 to 1aojA) Of the alignemnts tried so far (1pht, 1ckaA, 1bbzA, 1bb9, 1awj, 1aojA), the one that gets the best score for the sequence "foreign" sequence is 1ckaA/T0099-1ckaA-global.pw, which has 21 conserved residues and only two internal gaps (deletes P in one loop and inserts I in another). There is a minor N-terminal gap separating the first residue from the rest. The next best is 1bbzA/T0099-1bbzA-global.pw with 20 conserved residues, a 2-residue insertion and a 2-residue deletion. 1bb9/1bb9-T0099-fssp-global.pw has only 12 conserved residues and needs to delete two large pieces from 1bb9. (The alignment 1bb9/T0099-1bb9-vit.pw is similar.) 1ckaA/1ckaA-T0099-fssp-global.pw is almost the same as 1ckaA/T0099-1ckaA-global.pw, but moves the insertion to being the G next to the I, and closes up the N-terminus gap. This looks like the best alignment we're likely to get with this template. 1aojA/T0099-1aojA-global.pw is an interesting case---we get 18 conserved residues, an N-terminal gap, and a 1-residue insert. However, 1aojA is a dimer that forms two barrels when dimerized. This is quite different from the single barrel of the 1ckaA equivalence class. The link between the barrels is 1aojA leiLDDRRQwwk T0099 leiIEKEGDwwk which is precisely the region where 1ckaA has a proline missing in T0099. 1ckaA lriRDKPEEQwwn T0099 leiIEK-EGDwwk There is a strong possiblity that T99 is a dimer like 1aojA. 1awj/T0099-1awj-vit.pw has 15 conserved residues, with a one-residue insertion and a one-residue deletion. It looks ok, but no better than 1ckaA. The 1pht/T0099-1pht-vit.pw alignment has 18 conserved residues, with one big deletion. The highly conserved N-terminus looks good, but the deletion makes this an unsuitable template, compared to 1ckaA. It may be worth doing some cut-and-paste between different templates---taking the overall architecture from 1aojA, using the 1pht N-terminus and the 1ckaA C terminus. Mon Jun 26 09:52:23 PDT 2000 Remade 2ry predictions Mon Jul 24 14:21:27 PDT 2000 Kevin Karplus The big question for T99 is whether we give a monomeric or dimeric SH3 barrel (or two models, one of each). I have tried pasting together a chimeric dimeric model, using 1pht, 1ckaA, and 1aojA using undertaker, but undertaker doesn't understand dimers yet, so it may be no better than just doing a simple alignment to 1aojA. Mon Jul 24 Christian I think our two (very different) choice is between 1ckaA and 1aojA. I am slightly more inclined to go with 1aojA because its 1 isoleucine is probably more tolerable than the proline deletion in 1ckaA. In the latter case, there is not any "slack" to accomodate the deletion. Prolines often exist as rigid constraints for a conformation, thus I have the following conjecture: Looking at where the proline of 1ckaA would occur in 1aojA, it may be that the prolines presence or absence acts as a switch between the two conformations. 24 July 2000 Kevin Karplus It turns out that 1bu1[ADE] and 4hck are missing the proline also, but fold monomerically. Given the huge number of monomers and uniqueness of the dimer, I'm now more tempted to do a monomeric prediction. Also, since T99 is solved by NMR, perhaps the 5hck NMR template is the one to use. The T0099-5hck-all.pw.a2m alignment is gapless and has 23 conserved residues, so looks like our best bet. From karplus@cse.ucsc.edu Tue Jul 25 12:28:34 2000 Return-Path: Date: Tue, 25 Jul 2000 12:28:29 -0700 From: Kevin Karplus To: karplus@cse.ucsc.edu, rph@cse.ucsc.edu, saira@sanger.ac.uk, cline@cse.ucsc.edu, markd@cse.ucsc.edu, cbarrett@cse.ucsc.edu, stu@cse.ucsc.edu, rachelk@cse.ucsc.edu Subject: TS prediction for t99 I think I'd like to do an all-heavy-atom prediction for T99, based on the gapless alignment to 5hck. I have played around with undertaker, and I have reduced my choices to three. In pce/casp4/t99/und, the three choices are t99-scwrl.pdb t99-5hck-rot.pdb t99-5hck-opt2-rot.pdb The first two keep the backbone of 5hck, but set the rotamers using two different methods (SCWRL and undertaker). In the small tests I've done, SCWRL usually does slightly better than undertaker, but not always. The SCWRL assignment has one clash: CE1 of F2 and CD1 of I47, but there is plenty of room around there for the sidechains to wiggle away from each other. The undertaker-optimized rotamers have no clashes, and undertaker likes the wet spots, the dry spots, and the sidechain instrinsic scores better than it likes the SCWRL assignment. The third one has optimized the backbone and the rotamers using undertaker (with different scoring functions). Undertaker likes it a lot, but there is one clash (CB of F2 and CB of A54) that looks difficult to fix. The main change in the backbone is 54-56 which has peeled away from the beta sheet, and folded back on itself. (Using SCWRL on the same backbone gets 4 clashes). If I give both t99-scwrl.pdb and t99-5hck-rot.pdb to the biotech evaluation suite http://biotech.embl-heidelberg.de:8400/ it likes the SCWRL file much better. There are no unusual bond angles and only one unusual bond length (rather than 23 unusual angles and 6 unusual bond lengths for the t99-5hck-rot.pdb file). The conformation produced by undertaker has fewer buried polars (7 rather than 10), but more bumps (81 rather than 51). (Both SCWRL and undertaker are rather lax in their definition of clashes.) Most of the unusual angles are in GLU or LYS residues, probably because the undertaker library has a lot of rather poorly resolved sidechains in the training set. I should probably throw out residues from the training set with particularly high temperature or larger deviations from normal bond length or angle. Based on the WhatIF analysis, we're probably better off using the SCWRL'd chain than anything that undertaker did. Christian, can you format pce/casp4/t99/und/t99-scwrl.pdb for submission? The extra comments are just that we ran SCWRL on the alignment submitted in the AL format.