21 May 2000 Kevin Karplus target 93, HI0766 YIBK_HAEIN Hypothetical trna/rrna methyltransferase Nothing found with BLAST or double-blast, contradicting claim of "homologous sequence of known structure". But perhaps it is still close enough for PSI-BLAST and SAM-T99 to find. No PDB chain in T2K alignment of 74 sequences. Three fairly distinct groups of homologs---we may want to narrow in on a subfamily for more precise 2ry structure prediction. Best hit with target model: Top hits: E FSSP rep SCOP name 1d8cA 0.33 1d8cA 3.1.12 malate synthase g 1c8sA 18. 1c3wA 6.2.1 bacteriorhodopsin Best hits with template models: 1rkd 0.21 1rkd 3.67.1 ribokinase 1bhgA 3.5 1bhgA 2.1.4,2.17.1,3.1.7 beta-glucuronidase (gus gene product) 1a0p 28 1a0p 1.60.9,4.141.1 site-specific recombinase xerd 1ede 33 1b6g 3.64.1 Haloalkane dehalogenase 1glcG 34 1glcG 3.50.1 Glycerol kinase Tue Jun 6 13:35:30 PDT 2000 Redid 2ry prediction with new neural net. Mon Jun 12 14:53:22 PDT 2000 Looked at the CAFASP server predictions, which have fairly strong consensus: num SCOP 12 3.2.1 NAD(P)-binding Rossmann-fold domains 5 3.42.1 Thioredoxin fold 5 3.31.1 P-loop containing nucleotide triphosphate hydrolases 5 3.19.1 Nucleotidylyl transferase 4 4.132.1 N-terminal nucleophile aminohydrolases (Ntn hydrolases) Note: we do not get hits on any of these. Mon Jun 26 09:52:13 PDT 2000 Remade 2ry predictions Mon Jul 3 13:05:08 PDT 2000 Our secondary structure predictions do alternate helix and strand with 5 or 6 beta strands, so the Rossman-fold is a possibility. There are many fold families that are "rossman-like" 3.2, 3.4, 3.20-22, 3.25-26. With no strong hits, picking a possible template out of this large number of possibilities will be difficult. Going though some of our hits in order of score: The 1ede model scores well on reverse-sequence null, but not on simple null, so is unlikely. The alignment 1ede/1ede-T0093-global has some nice fragments (like some conserved turns), but the peices don't fit together into a compact structure. The 1rkd/1rkd-T0093-global alignment has a few fragments also, but lack the helix/strand alternation and is not really compact. (The 1rkd/1rkd-T0093-local alignment scores better with the simple null model, but is still missing enough that the 3D structure seems unlikely.) The TIM barrel in 1d8cA seems unlikely, since there are not enough beta strands. The best scoring of the alignments (1d8cA/T0093-1d8cA-vit) only matches one strand of the barrel. The 1d8cA/T0093-1d8cA-local alignment is larger, but no more convincing, as it is not compact in 3-space. The 1bhgA/1bhgA-T0093-vit and 1bhgA/1bhgA-T0093-local alignments are not convincing either, as they lack helices and are not compact. The gappy helices of 1c3wA/T0093-1c3wA-global are also unconvincing, as we have strong beta predictions for 2ry structure. This target may have to wait until we have the multi-track prediction working, so that we can use the 2ry structure prediction to help us select templates. I looked through the FSSP TABLE3 for methyltransferase, and found 52 hits. The FSSP representatives are 1af7 1.59.1,3.61.1 1bj4A 3.62.1 1booA 3.61.1 1mgtA 1.4.2+3.50.4 1sfe 1.4.2+3.50.4 1vid 3.61.1 1xvaA 3.61.1 2admA 3.61.1 2dpmA 3.61.1 2ercA 3.61.1 6mhtA 3.61.1 If we were making just a prediction just on function, then 3.61.1 looks promising. Since 2ercA is an rRNA methyltransferase, let's try is as a template. 2ercA/T0093-2ercA-global has a very bad gap, and almost no conserved residues. 2ercA/2ercA-T0093-vit picks up one helix and one edge strand. 2ercA/T0093-2ercA-vit picks up one helix and two strands. 2ercA/T0093-2ercA-local picks up one helix and one edge strand. 2ercA/2ercA-T0093-fssp-global scores terribly, but does not look as bad as the other alignments I've looked at for t93, but the secondary structure prediction seems to be way off, so the alignment is probably not so good. Should we try some of the other methyltransferases? Some of the Rossman folds? Tue Jul 11 11:39:35 PDT 2000 CAFASP consensus (weak) is 1qorA The alignment 1qorA/1qorA-T0093-fssp-global is not too bad---there is a missing final beta strand (unfortunately in mid-sheet), and the initial part of the sequence is mis-aligned (probably for lack of a FIM). The local alignment 1qorA/1qorA-T0093-local gets only a helix and one strand, and 1qorA/T0093-1qorA-vit gets a similar (slighlty longer) fragment. Mon Jul 17 Christian I have looked at FSSP neighbors around 1qorA: 1pedA 2ohxA 1fmtA 1qr6A 1a4iA None look promising. The FSSP methyltransferase representatives mentioned above didn't show anything exciting. The best was 2admA/2admA-T0093-vit.pw.dist:T0093 160 -8.12 -5.67 1.0e-02 1af7 3.61.1 1bj4A 3.62.1 1booA 3.61.1 1mgtA 3.50.4 1sfe 3.50.4 1vid 3.61.1 1xvaA 3.61.1 2admA 3.61.1 2dpmA 3.61.1 6mhtA 3.61.1 Thurs 20 July 2000 Kevin Karplus Rachel Karchin found a web site http://www.hwi.buffalo.edu/ACA/ACA00/abstracts/text/W0146.html with what appears to be information about both t92 and t93. Both are described as a "core methyltransferase fold of a central beta-sheet surrounded by alpha-helices on both faces." For t93, we have "a truncated version of the methyltransferase core domain, forming a tight dimer across a crystallographic symmetry axis mediated by two alpha-helices from each monomer." This means we probably want to look at 3.61 folds, assuming that we have found the right superfamily for t92. Friday 21 July Christian Reinvestigating the methyltransferase representatives above: 1af7/1af7-T0093-fssp-global.pw -2.59 -4.47 3.4e-02 Missing some core beta strands, but there a couple of seemingly interacting residue conservations 1vid/1vid-T0093-fssp-global.pw -3.76 -3.58 8.1e-02 With some hand editing, this may be a plausible prediction.