Weighing/filtering for local refinement

Submitted by bgreber on Thu, 2015-11-12 18:24

Forums

Frealign

Dear FREALIGN developers,

I am working on a complex that is composed of two relatively large masses that exhibit some motion relative to each other. Therefore, I aim to focus the alignment on one of them during refinement. I tried focused alignment in RELION using a mask to set to zero all the density values outside the mask. Because of the limitations of this approach, I tried to use the FREALIGN option to down-weigh and low-pass filter (instead of setting to 0) the parts of the complex outside the mask (e.g. weigh at 50% and filter to 20 Angstroms). However, this approach does not seem to improve the features of the region of interest compared to a refinement without the mask, as the lower overall resolution resulting from the application of the mask impacts the low-pass filtering of the output volumes (I run several cycles to allow the particles to move to their new optimum locations after mask application) and may also affect the sampling during refinement (though I am new to FREALIGN and am unsure if my interpretations are correct).

Do you have any recommendations regarding the use of the down-weighing/filtering feature for focused alignment? Are there any other parameters other than the weighing and filtering parameters that require adjustments?

Best regards,

Basil

niko Thu, 2015-11-12 18:41

The effect of the mask on the alignment will strongly depend on the geometry of your particle. Especially the low-resolution terms will keep your alignments more or less where they were without masking. To really see a difference and let the features inside the mask drive the alignment you may need to downweight the density outside the mask even further (e.g. 10%). At some point the weight will be so low that it is essentially like setting the density to 0. You probably have to try a few weights and maybe also a few different masks to find the optimum.

You may also try setting FBOOST to T (in mparameters). While this increases the chance of overfitting, as long as you keep the resolution limit of your refinement well below the resolution of your final reconstruction you can trust the resolution estimate from the FSC (more accurately the Part_FSC). FBOOST should help you get better alignments especially when you align against only a fraction of the density.

bgreber Thu, 2015-11-12 19:28

Thank you, Niko.

One more thought - would you recommend adjusting the parameter for the molecular weight of the particle to half if half of the particle mass is down-weighed to like 10-20% by the mask? I did not do that, as it is my understanding that the FSC calculation is performed on the unmasked output maps, and consequently, most of the particle mass will still be part of the reconstruction. Is this correct?

Thanks,

Basil

niko Thu, 2015-11-12 20:19

No, the mass should not be changed since the reconstruction will contain the full particle and should be filtered the same way as without a mask.