Pete, You are correct in assuming that what SHELXE calls 'weight' is what some other programs call FOM, i.e. an estimate of the mean cosine of the phase error. I have never had much confidence in the rather crude estimates of this number made by SHELXE, especially when more than about 20 cycles of density modification have been performed, although they are often in the right ball-park. However the FOM (and 'pseudo-free CC') values you obtained are unusually misleading. In general such examples of severe pseudo-translational symmetry often cause problems for experimental phasing.
George Prof. George M. Sheldrick FRS Dept. Structural Chemistry, University of Goettingen, Tammannstr. 4, D37077 Goettingen, Germany Tel. +49-551-39-3021 or -3068 Fax. +49-551-39-2582 On Sat, 8 Sep 2007, Dunten, Pete W. wrote: > Here's one - FOM > 0.5 after shelxe, SAD phases, not traceable > > Lots of weak data, as truncate shows . . . > > Analysis of mean intensity by parity for reflection classes > > For each class, Mn(I/sig(I)) is given for even and odd parity with respect > to the condition, > eg group 1: h even & odd; group 7 h+k+l even & odd; group 8 h+k=2n & h+l=2n & > k+l=2n or not > > 1 2 3 4 5 6 > 7 8 > h k l h+k h+l k+l > h+k+l h+k,h+l,k+l > > Totals: 22.0 22.9 21.9 23.0 39.4 5.3 39.4 5.3 21.9 23.0 22.0 22.9 > 22.4 0.0 37.9 17.0 > > The stats from shelxe for an incorrect solution > > Mean weight and estimated mapCC as a function of resolution > d inf - 2.43 - 1.93 - 1.68 - 1.52 - 1.41 - 1.33 - 1.26 - 1.21 - 1.15 - > 1.06 > <wt> 0.581 0.625 0.614 0.610 0.603 0.594 0.600 0.583 0.574 0.547 > <mapCC> 0.988 0.981 0.979 0.980 0.978 0.979 0.978 0.975 0.969 0.955 > N 4117 4037 4172 4272 4177 3923 4338 3632 4456 3638 > > Pseudo-free CC = 75.58 % > > and the other hand > > Mean weight and estimated mapCC as a function of resolution > d inf - 2.43 - 1.93 - 1.68 - 1.52 - 1.41 - 1.33 - 1.26 - 1.21 - 1.15 - > 1.06 > <wt> 0.589 0.635 0.617 0.622 0.623 0.621 0.625 0.610 0.602 0.579 > <mapCC> 0.987 0.982 0.981 0.982 0.980 0.983 0.982 0.978 0.976 0.968 > N 4117 4037 4172 4272 4177 3923 4338 3632 4456 3638 > > Pseudo-free CC = 78.72 % > > and the correct solution, which was traceable > > Mean weight and estimated mapCC as a function of resolution > d inf - 2.43 - 1.93 - 1.68 - 1.52 - 1.41 - 1.33 - 1.26 - 1.21 - 1.15 - > 1.06 > <wt> 0.571 0.623 0.612 0.621 0.623 0.618 0.623 0.608 0.615 0.605 > <mapCC> 0.987 0.984 0.983 0.984 0.982 0.984 0.983 0.980 0.980 0.975 > N 4117 4037 4172 4272 4177 3923 4338 3632 4456 3638 > > Pseudo-free CC = 85.42 % > > Pete > > > -----Original Message----- > From: CCP4 bulletin board on behalf of James Holton > Sent: Sat 9/8/2007 8:57 AM > To: CCP4BB@JISCMAIL.AC.UK > Subject: Re: [ccp4bb] just how bad can phases be and still help > > In my experience, the "threshold of interpretability" for electron > density maps is when the FOM is ~0.5 or higher. I am basing this on > anecdotal responses to this movie: > http://bl831.als.lbl.gov/~jamesh/movies/index.html#phase > In this case, the displayed value for FOM is exact. Depending on how > you are estimating FOM, your mileage may vary. > > As noted on another thread, different programs do indeed report > different values for the same statistic. The reason why different > programs give different FOMs is because the definition of FOM is the > cosine of the difference between the reported phase (output by the > program) and the "true" phase. Since you obviously don't know the > "true" phase, this is a hard number to figure out. Every phasing > technique has a different method of estimating the FOM, but all of them > are just that: estimates. In fact, it is somewhat preposterous to try > to compute the deviation from some unknown "correct" value, which is why > we have so many other phase quality statistics like phasing power, lack > of closure, RCullis etc. These are at least well-defined for a given > data set, but unfortunately don't have much more than an empirical > correlation to the number you really want to know: FOM. > > Map interpretability depends on the "true" FOM, but unfortunately, all > we have to work with is the FOMs we can estimate. There are sometimes > practical reasons for underestimating the FOM (see the mlphare manual), > and a solvent-flattening job that is allowed to run for far too many > cycles will get phase bias and overestimate the FOM. I can say that > in controlled tests I have done, the FOM reported by dm (if you have a > modern version and let it decide when to stop cycling) tends to be > fairly accurate. Yes, it does depend on resolution, but if you have > high-res data with very low FOM, then you basically don't have any > high-res data in an FOM-weighted map. I tend to predict map > interpretability by asking the question "to what resolution limit is the > average FOM > 0.5?" If that is 10A, then I generally don't bother > looking at the map. > > Perhaps the best way to settle this is to put it out to a challenge: > does anybody have a map calculated with an average FOM(fromsomeprogram) > < 0.5 that they were able to trace? Does anyone have a map calculated > with FOM(fromsomeprogram) > 0.5 that was total garbage? > > -James Holton > MAD Scientist > > > Bryan W. Lepore wrote: > > general question - perhaps the fundamental question - > > > > for anyone who had "weak/poor/bad" phases from some source, that were > > later actually used to solve a structure when combined w/ another > > source - HOW bad were the worst phases on their own, in terms of > > resolution, FOM, CC, e-density, (any other numbers)? what was MOST > > important in knowing the phases would help (presumably e-dens.). > > > > i.e, was it only when relatively "better" phases gave any > > interpretable density that it was known that the "bad" phases would help? > > > > -bryan >
