Thanks all for the insights. Both of the suggestions are very helpful.
Best,
Sandip
On Thu, Mar 8, 2018 at 4:18 AM, Sjors Scheres <[log in to unmask]> wrote:
> Dear Sandip,
>
> I would recommend fitting the atomic model in one of the half-maps from
> a gold-standard refinement. And then calculate the FSC between that
> refined model and that same half-map (FSC_work), PLUS calculating the
> FSC between that same model and the OTHER half-map (FSC_test). A large
> difference between these 2 curves are an indication of overfitting.
> Overfitting result in unrealistic models (with worse geometrical
> statistics) and over-estimated FSC curves, and should be avoided. You
> can limit overfitting by setting a higher weight on the geometric terms
> in your fitting. Therefore, you could repeat this procedure with
> different weights to assess where overfitting starts to happen. Then,
> you could choose the smallest weight where it doesn't happen yet and
> perform your final refinement of the model with that weight in the
> combined map from both halves. Examples of this are in Amunts et al
> (2014) (http://dx.doi.org/10.1126/science.1249410), or in more detail in
> Fernandez et al (2014) (http://dx.doi.org/10.1016/j.cell.2014.04.015).
>
> HTH,
>
> Sjors
>
>
>
> On 03/07/2018 09:32 PM, Sandip Basak wrote:
>> Hi All,
>>
>> I have a situation and I have come to a conclusion. I am doing real space refinement in phenix. When I don't specify weight in phenix I get good fitting (CC_mask: 0.8118) but the Ramachandran plot shows 88.64% favored and 11.36% allowed. If I specify weight (0.1 or other) I see fitting become worse (CC_mask: 0.6918) but the Ramachandran plot shows 91.52% favored and 8.48% allowed. Now which strategy is good? Should I put weight or let the phenix decide the weight?
>>
>> Thanks in advance
>>
>> Best
>> Sandip
>
> --
> Sjors Scheres
> MRC Laboratory of Molecular Biology
> Francis Crick Avenue, Cambridge Biomedical Campus
> Cambridge CB2 0QH, U.K.
> tel: +44 (0)1223 267061
> http://www2.mrc-lmb.cam.ac.uk/groups/scheres
>
|