Adaptive bandpass filtering

From Dynamo
Jump to navigation Jump to search


The "golden standard" methodology is a widespread method to drive the alignment process avoiding overfitting and providing an accepted estimation of the attained resolution.

It bases on splitting the original data set into two equally sized halves (called even/odd), and aligning them separately. After each iteration, the attained resolution is estimated by comparing the two separately computed averages through FSC. The bandpass filter used in the next iteration will be defined based on this estimation.


Operationally, an ABP project is just a multireference project that:

  • doesn't swap particles between references, and
  • and that adjust the lowpass filter parameters at each iteration.

The most comfortable way to create an ABP project is by first designing a single reference project (i.e myProject), and then letting Dynamo derive an ABP project, which will be generally called myProject_eo.

The ABP procedure workflow

Initialization

The data will be split in two halves, and separate initial templates will be computed. An initial frequency needs to be explicitly set.

Iteration

Each iteration proceeds normally for each half data set. After computation of the two averages, they might be aligned to each other (to allow a meaningful comparison) and symmetrized.

When computing the FSC, the direct space smoothing mask passed by the user will be applied to each one of the half averages.

Parameters in an ABP project

When the parameter adaptive_bandpass is switched on, Dynamo is triggered to perform the resolution analysis incaranted by the ABP procedure.

Several parameters can be chosen by the user. When designing the project from the dcp GUI, they are available in Numerical parameters > General


Initial frequency

Given in Fourier pixels. It is passed to the project just as the low frequency bandpass of the first round. From the first iteration on, the bandpass values indicated for the different rounds will be ignored, as Dynamo will recompute them.

Alternatively, you can also set the parameter adaptive_bandpass_initial". If this parameter is not zero, it will overrun the value indicated in the regular rounds.

Alignment of half averages

The alignment of the two half averages proceeds by keeping half average 1 fixed and rotating and shifting half average 2 searching for a cross correlation maximum. The angular parameters for this search are the same ones used during the iteration itself.

Switching on the parameter adaptive_bandpass_symmetrized, the resolution estimation will be performed on the symmetrized half averages (in case the round) has a nontrivial symmetrization parameter.

Resolution estimation

FSC threshold

You can choose the FSC value that defines the resolution estimation with adaptive_bandpass_threshold. The default is 0.5, typically you should change it to 0.143.

Resolution pushback

"Pushing back" the resoution means being conservative: if Dynamo finds that your resolution threshold (say FSC = 0.143) is attained at 15 fourier pixels, then a pushback of 2 means that the bandpass used in the iteration will be set to 13.

Creating an ABF project in dcp

In general, you would design a single reference project. Then you would derive the full ABP project. When your single reference project is on scope in the 'Project' field, just clicking on the tabs on the top row of dcp "Multireference > Adaptive Filtering (golden standard) > Derive project". The new ABP project created by Dynamo will be put on scope in the dcp_GUI. It will inherit the alignment parameters and linked files (mask, data, etc) of the original project, but will be a multireference project set to perform the ABP procedure. Multireference means that you will see two reference channels, but without particle swapping

Creating an ABF project in the command line

[Sorry, not yet ready!]