[phenixbb] Unmasking cavities

[Pavel Afonine]

Hi Morten,

> I get the "add dummy atoms and calculate map" to check whether it is 
> Fourier truncation ripples (which I don't think it will turn out to be).
> But I wouldn't feel comfortable depositing a structure with dummy 
> atoms even if they do have zero occupancy. Are you really suggesting 
> that people do that?

Of course not. The above suggestion was to test the idea about where the 
negative peaks are coming from (just to mask the region in order to 
avoid placing a flat bulk solvent density there and see if the peaks are 
still there).

Just for your information: there is a number of files in PDB that 
contain "dummy atoms" which is obviously bad (since scattering type is 
undefined which makes it impossible to compute the R-factors for such 
models). Here are a few examples: 2aed, 2ajp, 2amx, 2ar0, 1au9, 2awp, 
2ayv, 2b25, ... I can go on, I have a complete list...

> Secondly, when I look in the .def for my refinements I find two 
> entries for mask calculation:
> Under the fake_f_obs heading
>     mask {
>       solvent_radius = 1.11
>       shrink_truncation_radius = 0.9
>       grid_step_factor = 4
>       verbose = 1
>       mean_shift_for_mask_update = 0.1
>       ignore_zero_occupancy_atoms = True
>       ignore_hydrogens = True
>   }
> And again under it's own heading towards the end
>   mask {
>     solvent_radius = 1.11
>     shrink_truncation_radius = 0.9
>     grid_step_factor = 4
>     verbose = 1
>     mean_shift_for_mask_update = 0.1
>     ignore_zero_occupancy_atoms = True
>     ignore_hydrogens = True
>   }
>
> Which one is relevant? 

Both are relevant, but used for different purposes.
The first one is used for experimenting/development/testing ideas, when 
real Fobs are replaced with calculated ones (as the scope name suggests: 
fake_f_obs):
fake_Fobs = Fmodel = scale_k1 * exp(-h*U_overall*ht) * (Fcalc + k_sol * 
exp(-B_sol*s^2) * Fmask)

The second scope defines the mask parameters for everything else and 
this is the one that you most likely want to play with.

> Also why didn't any of you suggest the optimize_mask=true parameter? 
> Shouldn't that automatically find the best solvent_radius and 
> shrink_truncation_radius values?

It's a good idea to try out, but that mask optimization does a grid 
search over a range of solvent_radius and shrink_truncation_radius and 
the "best" values are chosen based on the lowest R (or Rfree, I don't 
remember exactly). I wouldn't believe that fixing a few negative peaks 
somewhere will show up in R-factor (R-factor is a global measure and 
very insensitive to minor local changes).

Cheers,
Pavel.