With a long axis, the twinning test often falsely suggest twinning, as the spots may not be properly resolved. Thus a weak reflection may be overestimated because it is next to a strong one, leading to the same intensity statistics that you get in twinning
Phil > On 6 Dec 2016, at 14:49, Eleanor Dodson <[email protected]> wrote: > > The twinning factors all about 0.25 do point to a higher symmetry than P21. > As others say - I would be very surprised (and would question result as a > referee) if your FreeR was not >> 30% for such low resolution data > > Is the data anisotropic - often is with such a long cell edge.. > If so do process to the highest possible resolution in the a b orientation - > the quality indicators will be much higher with aniso data and you may be > throwing away good measurements . > > Eleanor > > On 6 December 2016 at 14:37, Andreas Forster <[email protected]> wrote: > Dear Andre, > > I agree with Jacob that P 42 21 2 might be the right space group, and that > and R free of 33% isn't so bad. If your electron density is poor, that might > just reflect the low resolution. Have you tried refinement with Buster? It > has a way with low-resolution data. > > All best. > > > Andreas > > > > On Tue, Dec 6, 2016 at 3:25 PM, Keller, Jacob <[email protected]> > wrote: > What were the twinning tests like in p42212? I suspect that it is really that > spacegroup, the low Rfree is artificial (33% is not too bad at 3.9 Ang) and > the electron density is bias. > > > > JPK > > > > From: CCP4 bulletin board [mailto:[email protected]] On Behalf Of Andre > Luis Berteli Ambrosio > Sent: Tuesday, December 06, 2016 9:17 AM > To: [email protected] > Subject: [ccp4bb] Choosing test set for twin refinement, with multiple > operators > > > > Dear all, > > We are currently refining a low resolution model (3.9 A max), obtained by MR. > Dataset was collected from a single crystal with one long cell axis (~620 A) > and high solvent content (74%). > > Best refinement results (by far!) are obtained in Refmac, by imposing P21 sg > (20 multidomain monomers/AU) and allowing for amplitude-based twin refinement. > > Accordingly, Refmac identifies four twin operators, which I understand have > considerable fractions, as follows: > > > > **** Twin operators with estimated twin fractions **** > > > > Twin operator: H, K, L: Fraction = 0.249; Equivalent operators: -H, K, -L > > Twin operator: -H, -K, L: Fraction = 0.260; Equivalent operators: H, -K, -L > > Twin operator: -K, -H, -L: Fraction = 0.235; Equivalent operators: K, -H, L > > Twin operator: K, H, -L: Fraction = 0.257; Equivalent operators: -K, H, L > > > > I suspect that the low Rfree (~25%) may be artificial and results from an > inappropriate selection of the test due to the multiple operators. > > However, I cannot figure out how to properly select the test set when such a > high number of operators must be considered. > > FYI, space group P42212 resulted in the best processing statistics (5 > monomers/AU); however, model never improved from an Rfree of 33%, with the > quality of the electron density distribution for some of the domains being > heavily compromised. > > I honestly apologize if I am missing some obvious points here and will be > happy to provide more info, including statistics on data processing. > > Since is a multidomain protein, is it also appropriate to include TLS > refinement at this low resolution? > > I thank you all in advance. > > Best, > > Andre. > > > > > > > > > > >
