[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [ccp4bb]: REFMAC5 maximum likelihood refinement : R increased



Dear Wolfram,

under which circumstances would maximum likelihood target refinement in
REFMAC5 produce a dramatic INCREASE in R (from 0.187 to 0.208) and Rfree
(from 0.232 to 0.257) with 1.8A data? This was not accompanied by
improved geometry (reduced RMSs). Any idea where I should start looking
for the underlying problem?
Thank you.
Wolfram Tempel

When you do ML refinement you "push" the calculated structure factor magnitudes NO LONGER to Fobs (as it is the case for LS refinement) BUT to some MODIFIED values F*.

        Essentially, the main "trick" behind ML is that it takes into account the missed part of your model. Instead of fit
        Fmodel --> Fobs
you start to fit
        Fmodel + Fmissed --> Fobs
(Fmissed are estimated statistically from ML; such modification allows one also to take into account other imperfections in you current model).

        From this you see that your Fmodel theoretically can start to diverge from Fobs, especially if your model is imperfect , e.g. is incomplete enough (if Fmissed are significant).

        For some more detail, you can look, for example, Lunin, Afonine & Urzhumtsev (2002) "Likelihood-based refinement. 1. Irremovable model errors.". Acta Cryst., A58, 270-282.

Regards,

Sasha