I have tested this a lot and I get the same result in the sense that after extensive refinement there is no evidence of bias. That is, the R-free from previously used test sets is not systematically lower than that of the original 'free' test set. I used REFMAC for that. In the context of PDB-REDO, I reset the B-factor to a flat value related to the Wilson B-factor and then do many cycles of refinement. That would be whatever I used before in PDB-REDO given the kind of refinement needed for the model + 40 (see 10.1107/S0907444911054515). I only add coordinate noise for cases where the model already has a flat B-factor (very low resolution cases mostly), but very little (gemmi set --noise 0.05). Adding too much noise makes a mess and to be honest haven't really seen that the noise adds that much except that some referees are particularly keen on it.
Cheers, Robbie > -----Original Message----- > From: CCP4 bulletin board <CCP4BB@JISCMAIL.AC.UK> On Behalf Of Tim > Gruene > Sent: Wednesday, June 25, 2025 09:49 > To: CCP4BB@JISCMAIL.AC.UK > Subject: Re: [ccp4bb] free R in shells > > On Tue, 24 Jun 2025 22:54:28 +0100 > Martin Malý <martin.maly...@email.cz> wrote: > > > Actually I would be quite interested what is your best practise when > > you want to reset a structure model to not be biased towards any test > > reflection set. > Dear Martin, > > according to Ian Tickle, the idea or 'resetting' the structure is the wrong > understanding, and if you notice model bias, there was a mistake in the > approach. You can safely build your model and refine against all data all the > way to the end. When you are interested in Rfree, you pick any random set > and refine to convergence, i.e. until the target function does not drop any > further. Refinement removes bias. (see Ian's emails on this topic, quite some > years back) > > 'shaking' the model coordinates and/or adps is no good. I remember in one > case when I worked on Rcomplete, there was a side chain the flipped into a > different rotamer from shaking. This is a different model, and you cannot > average R-values from different models (see sections "Effect of Parameter > Perturbation." and "Influence of Parameter Perturbation on Convergence > Rate" and Figures S4 /S5 in > https://doi.org/10.1073/pnas.1502136112 / > https://www.pnas.org/doi/suppl/10.1073/pnas.1502136112/suppl_file/pnas. > 1502136112.sapp.pdf > > Best, > Tim > > > > -- > -- > Tim Gruene > Head of the Core Facility Crystal Structure Analysis Faculty of Chemistry > University of Vienna > > Phone: +43-1-4277-70202 > > https://ccsa.univie.ac.at > > GPG Key ID = A46BEE1A > > ############################################################### > ######### > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 > > This message was issued to members of www.jiscmail.ac.uk/CCP4BB, a > mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at > https://www.jiscmail.ac.uk/policyandsecurity/ ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 This message was issued to members of www.jiscmail.ac.uk/CCP4BB, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
