Multireference alignment

From Dynamo
Jump to navigation Jump to search

In a multireference alignment [MRA] project, each particle will be aligned against R different templates. Multireference projects are mainly used to incarnate the Multireference analysis approach for simultaneous alignment and classification.

The process of multireference alignment

If you have R references and N particles, each particle will be aligned against all the R references. The particle will produce R different cross-correlation scores and R different sets of alignment pararameters. Thus, Dynamo will produce R different refinement tables, each for one reference channel.

Each particle will thus appear on all tables, but it will yield a maximal cross-correlation score in one of them... informally, the particle chooses that reference in the current iteration. When computing the averages that will serve as templates for the next iteration, only those particles that "choose" a given reference will contribute to the average in that reference channel.


The fact that you can define alignment masks and classification masks separately in a project is specially useful when running MRA projects. Very frequently, you want to use a wide mask that includes enough signal to drive the alignment, and a smaller mask, defined on the region where you expect or want to check for structural heterogeneity.

Input format

Introducing project elements (template, tables) for MRA projects is slightly different than introducing them for single reference projects. You can use a .sel file, or pass a folder, where individuals files need to stick to a given naming convention. Each reference number is given by a three figure zero padded integer starting with zero.

<folder seeds>/template_initial_ref_001.em
<folder seeds>/fmask_initial_ref_001.em
<folder seeds>/table_initial_ref_001.em

Here, <folder seed> is the name of a folder that can be passed to a project as parameter for template instead of the name of a file. Same holds for parameters table and fmask. You can use different folders for different parameters.


The dcp GUI gives you the option of just cloning a previously available table.

Initial templates

There are different policies for the creation of the first templates.

A priori information

In some circumstances, you can use geometrical shapes as initial templates to drive the classification.

Adding noise to a given map

The dcp GUI gives the option of using a template already available in the GUI and producing multiple copies by adding different realizations of gaussian noise of the same amplitude.

Random subset averaging

You can use an avialable table and data folder and construct averages taking N-different subsets of particles. This can be performed with an order like:

dynamo_write_multireference(myTable,'table',myFolder,'refs',1:4,'data',myData,'subset',M/4); will create 4 subsets of particles each containing M/4 particles (M being the number of valid particles in the table), produce 4 averages, store the files in myFolder and format the individual file names so as to follow the appropriate convention


You will get a different refined table and iteration average for each reference channel and iteration.

Fast command line access

The ddb command is probably the easiest way of accessing the different results in a multireference alignment .

Swapping history

It is usually a good idea to check how the particle membership evolves during the iteration procedure. A healthy multireference alignment registers many particle swaps between reference channels in the first iterations. This swapping should decrease gradually until the different channels just evolver independently. This can be checked through the GUI with dcp GUI > show > Multireference