Re: [ccp4bb] molecular replacement and twinning problems
Well - there s nothing there to indicate twinning, although as you say it is hard to be sure with a pseudo translation... The contrast doesnt look brilliant, but then it often doesnt.. What about phaser? If it gives a contrast between one spacegroup and the others I always think that is a good sign.. It will take a while though.. Eleanor Phil Evans program othercell suggests you could reindex to get two axes equal.. And that is a prequisite for twinning . C 1 2/m 1 192.3 192.3 117.2 90.0 90.0 92.01.98 [k-l,k+l,h] Same cell [-k+l,-k-l,h] Eleanor Kay Diederichs wrote: Eleanor Dodson schrieb: You dont mention any twinning tests? sorry, I forgot to mention that the twinning tests do not show twinning. Rather, the actual curves in the "Cumulative distribution of H" lie on the "not-twinned-at-all" (i.e opposite) side of the alpha=0 curve (see plot below). But I'm pretty sure that this is due to the pseudo-translation (almost centering) which results in a high proportion of very weak reflections - contrary to what you get with twinning. That's what I get from sfcheck, when run in P212121: Perfect twinning test /^2 : 3.1699 Partial Twinning test: -h,+l,+k Polar angles: 135.00 -89.99 179.99 Alpha(twin fraction),Npair,Ior,Tol :-0.109 1625882 0.030 --- Partial Twinning Test : H = !I(h1)-I(h2)!/(I(h1)+I(h2)) --- Alpha(twinning fraction) = 1/2 - Reflection related to hkl : -h,+l,+k Cumulative distribution of H 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 *---+---+---+---+---+---+---+---+---+---+ O+++.+ . . . . . . . . ! O ++.+ . + . . . . . . . ! !O +++ + . +. . . . . . ! 0.10!-O-++-+--+-+---! ! O .++ + .+ . . .+ . . . ! ! O. + + + . +. . . . + . . ! ! O +.+ + . + . . . . +. ! 0.20!---O---+-+--+--+---+ alpha=0.4 ! .O .+ +. +. . +. . . . + ! . O . + + .+ . .+ . . . + ! . O . +. + . + . . + . . + 0.30!--O+--++-+-+ ! . O .+ + . + . . .+ . + ! . O . + .+ . +. . . +. + ! . .O . +. +. .+ . . . + + 0.40!-O-+---+--++ alpha=0.3 ! . . O. .+ .+ . + . . + ! . . O. . + . +. . + . . + ! . . O . +. + . + . + 0.50!O--++---+--+ ! . . .O . .+ . + . . +. + ! . . . O . . + . + . .+ + ! . . . O. . +. .+ . . + + 0.60!---O---+-+-+ alpha=0.2 ! . . . .O . .+ . +. . + ! . . . . O . . + . .+ . + ! . . . . O. . +. . + . + 0.70!--O+--++ ! . . . . O . .+ . + + ! . . . . .O . . + . .+ + ! . . . . . O. . +. . ++ 0.80!---O---+---+ alpha=0.1 ! . . . . . .O . .+ . + ! . . . . . . O . . + . + ! . . . . . . O. . +. + 0.90!O--+---+ ! . . . . . . . O. .+ + ! . . . . . . . .O . + + ! . . . . . . . . O. ++ H *---+---+---+---+---+---+---+---+---+---* alpha=0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Of course this plot looks _very_ different if I run it in P1 because then sfcheck uses -h,k,-l as twinning operator - it then looks like perfect twinning. The L test, now part of the newest ctruncate is pretty good at detecting twinning even with the NCS translation. this is the L test from ctruncate 1.0.0 : 30/10/08 run in P212121 : $TABLE: L test for twinning: $GRAPHS: cumulative distribution function for |L|:0|1x0|1:1,2,3,4: $$ |L| Observed Expected_untwinned Expected_twinned $$ $$ 0.00 0.00 0.00 0.00 0.05 0.048142 0.05 0.074938 0.10 0.093721 0.10 0.149500 0.15 0.139205 0.15 0.223312 0.20 0.184799 0.20 0.296000 0.25 0.230867 0.25 0.367188 0.30 0.277173 0.30 0.436500 0.35 0.323650 0.35 0.503563 0.40 0.370747 0.40 0.568000 0.45 0.418233 0.45 0.629437 0.50 0.466569 0.50 0.687500 0.55 0.515615 0.55 0.741812 0.60 0.565706 0.60 0.792000 0.65 0.616860 0.65 0.837688 0.70 0.669232 0.70 0.878500 0.75 0.723200 0.75 0.914062 0.80 0.779144 0.80 0.944000 0.85 0.836994 0.85 0.967938 0.90 0.896933 0.90 0.985500 0.95 0.957486 0.95 0.996313 1.00 1.00 1.00 1.00 $$ And SFCHECK does a good job too. If these are inconclusive I would not assume twinning. Usually you can get solutions for MR with twinned data, b
Re: [ccp4bb] molecular replacement and twinning problems
Eleanor Dodson schrieb: You dont mention any twinning tests? sorry, I forgot to mention that the twinning tests do not show twinning. Rather, the actual curves in the "Cumulative distribution of H" lie on the "not-twinned-at-all" (i.e opposite) side of the alpha=0 curve (see plot below). But I'm pretty sure that this is due to the pseudo-translation (almost centering) which results in a high proportion of very weak reflections - contrary to what you get with twinning. That's what I get from sfcheck, when run in P212121: Perfect twinning test /^2 : 3.1699 Partial Twinning test: -h,+l,+k Polar angles: 135.00 -89.99 179.99 Alpha(twin fraction),Npair,Ior,Tol :-0.109 1625882 0.030 --- Partial Twinning Test : H = !I(h1)-I(h2)!/(I(h1)+I(h2)) --- Alpha(twinning fraction) = 1/2 - Reflection related to hkl : -h,+l,+k Cumulative distribution of H 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 *---+---+---+---+---+---+---+---+---+---+ O+++.+ . . . . . . . . ! O ++.+ . + . . . . . . . ! !O +++ + . +. . . . . . ! 0.10!-O-++-+--+-+---! ! O .++ + .+ . . .+ . . . ! ! O. + + + . +. . . . + . . ! ! O +.+ + . + . . . . +. ! 0.20!---O---+-+--+--+---+ alpha=0.4 ! .O .+ +. +. . +. . . . + ! . O . + + .+ . .+ . . . + ! . O . +. + . + . . + . . + 0.30!--O+--++-+-+ ! . O .+ + . + . . .+ . + ! . O . + .+ . +. . . +. + ! . .O . +. +. .+ . . . + + 0.40!-O-+---+--++ alpha=0.3 ! . . O. .+ .+ . + . . + ! . . O. . + . +. . + . . + ! . . O . +. + . + . + 0.50!O--++---+--+ ! . . .O . .+ . + . . +. + ! . . . O . . + . + . .+ + ! . . . O. . +. .+ . . + + 0.60!---O---+-+-+ alpha=0.2 ! . . . .O . .+ . +. . + ! . . . . O . . + . .+ . + ! . . . . O. . +. . + . + 0.70!--O+--++ ! . . . . O . .+ . + + ! . . . . .O . . + . .+ + ! . . . . . O. . +. . ++ 0.80!---O---+---+ alpha=0.1 ! . . . . . .O . .+ . + ! . . . . . . O . . + . + ! . . . . . . O. . +. + 0.90!O--+---+ ! . . . . . . . O. .+ + ! . . . . . . . .O . + + ! . . . . . . . . O. ++ H *---+---+---+---+---+---+---+---+---+---* alpha=0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Of course this plot looks _very_ different if I run it in P1 because then sfcheck uses -h,k,-l as twinning operator - it then looks like perfect twinning. The L test, now part of the newest ctruncate is pretty good at detecting twinning even with the NCS translation. this is the L test from ctruncate 1.0.0 : 30/10/08 run in P212121 : $TABLE: L test for twinning: $GRAPHS: cumulative distribution function for |L|:0|1x0|1:1,2,3,4: $$ |L| Observed Expected_untwinned Expected_twinned $$ $$ 0.00 0.00 0.00 0.00 0.05 0.048142 0.05 0.074938 0.10 0.093721 0.10 0.149500 0.15 0.139205 0.15 0.223312 0.20 0.184799 0.20 0.296000 0.25 0.230867 0.25 0.367188 0.30 0.277173 0.30 0.436500 0.35 0.323650 0.35 0.503563 0.40 0.370747 0.40 0.568000 0.45 0.418233 0.45 0.629437 0.50 0.466569 0.50 0.687500 0.55 0.515615 0.55 0.741812 0.60 0.565706 0.60 0.792000 0.65 0.616860 0.65 0.837688 0.70 0.669232 0.70 0.878500 0.75 0.723200 0.75 0.914062 0.80 0.779144 0.80 0.944000 0.85 0.836994 0.85 0.967938 0.90 0.896933 0.90 0.985500 0.95 0.957486 0.95 0.996313 1.00 1.00 1.00 1.00 $$ And SFCHECK does a good job too. If these are inconclusive I would not assume twinning. Usually you can get solutions for MR with twinned data, but I havent much experience of the signal quality.. Can you solve it in P1 then sort out the spacegroup later? I tried; this is the full story: I ran molrep (version 10.2.12 from CCP4 6.1.0) in P1, with NMON=8. It uses the pseudo-translation vector and thus places 4 times 2 molecules. In the "fast" mode (standard RF and TF without rigid body refinement) the "contrast" is 1.93/1.77/1.87/14.72 for the 1st/2nd/3rd/4th pair of molecules. However the result is different if I run it in "medium" (contrast=2.82/4.56/1.55/3.10) or "slow" (2.87/11.47/2.01/2.59) mode, and molrep only writes out 4 molecules instead of 8, in these two modes. I therefore s
Re: [ccp4bb] molecular replacement and twinning problems
You dont mention any twinning tests? The L test, now part of the newest ctruncate is pretty good at detecting twinning even with the NCS translation. And SFCHECK does a good job too. If these are inconclusive I would not assume twinning. Usually you can get solutions for MR with twinned data, but I havent much experience of the signal quality.. Can you solve it in P1 then sort out the spacegroup later? Eleanor Kay Diederichs wrote: Dear all, we have crystals that nicely diffract to 1.7 A (sharp spots), with the following characteristics and findings: a) the data appear as P212121, with axes 117.2 133.6 138.3 (if reduced in P1, the largest deviation of any angle from 90° is 0.2°); the odd screw-axis reflections are mostly indistinguishable from noise; the data do not scale significantly better in P2/P21 (any setting) or P1. b) there is a good model available, with coords known from a complex of this protein with another one; two molecules of this model would give 64% solvent in P212121 which appears reasonable for a membrane protein c) the structure cannot be solved with this model in P212121, nor can it be solved in P222, P2122, P2212, P2221, P21212, P22121, P21221 d) we conclude that the true space group must be P2 or P21 (with one of the three possible settings), with almost-perfect twinning. Or it is P1 with tetartohedral twinning. There are thus still six + one possibilities. e) MOLREP tells us --- Check Patterson for pseudo-translation --- PST_limit : 0.125 of origin peak INFO: pseudo-translation was detected. Origin Patterson peak: P,P/sig :57.748 257.690 1 Patterson. peak: p,P/sig :28.773 128.395 2 Patterson peak : P,P/sig :16.55173.856 3 Patterson peak : P,P/sig : 8.50237.936 Peak 1: trans.vector /ort/ : 0.01155.68869.399 trans.vector /frac/: 0.000 0.416 0.500 Peak 2: trans.vector /ort/ :58.55466.863 0.000 trans.vector /frac/: 0.500 0.500 0.000 Peak 3: trans.vector /ort/ :58.56511.38569.399 trans.vector /frac/: 0.500 0.085 0.500 INFO: translation vector of peak 1 will be used. Two molecules (for the orthorhombic spacegroups) may produce only one pseudo-translation vector. As there is more than one strong pseudo-translation vector, I conclude that we have at least 3 molecules in the ASU (consistent with monoclinic). f) we've calculated all seven molecular replacement searches of d) in MOLREP. The contrast is very high in all cases. However, Refmac rigid-body refinement of the solutions, with "Twin refinement" activated, gives about 51% R and the same for Rfree (give or take 1 %), in all cases. I'm wondering: how reliable is a rotation search in the presence of perfect twinning? Is there any molecular replacement program that can take a given twinning operator into account in the rotation and translation search? Any other hints what to try? best, Kay
