Vesicle models

From Dynamo
Jump to: navigation, search

Vesicle models are appropriate when particles are evenly distributed on surfaces reasonably similar to spheres or ellipsoids. The model can estimate the center and radius provided points provided by the user in the visible part of the object, or the other way round, letting the user provided center and radius.

Important parameters

Radius

Given in pixels. Can be estimated or provided by the user.

Center

Given in pixels. Can be estimated or provided by the user.

Separation

Given in pixels. Defines the mean separation of the cropping positions to be defined in the vesicle surface. When choosing this value, you should think on oversampling the actual distribution of particles that you expect.


Picking vesicles in a tomogram browser

There are different methods for vesicle picking interactively, depending on the precision you need. When choosing your method, remember that, if necessary, vesicles can be fitted against the tomogram density map in a later stage. If such a fitting is possible, you don't need to spend too much time trying to make your manual picking too precise, but just reasonable enough to be used as a seed for the automated fitting procedure.

Cloud of points

The most robust method requires from the user to pick several points in each vesicle model, and then let Dynamo fit the radius and the center. This procedure can be performed through the browser dtmslice, as shown in the next images.


A tomogram with several vesicles shown in dtmslice

First of all we need to define a model of type vesicle into the model pool viewed by dtmslice.

Creation of a vesicle model

Once the model is created an active, we can navigate in the tomogram (for instance, by dragging the slice through movement of the cursor while the main mouse button is pressed). Three layers are normally enough.

Click vesicle points (in pink) on different z levels (use [c] to pick each)

After collecting four of five points on each layer, we can zoom out (with, for instance, [-] ) and bring the slice to a slower z value.

View from a lower z level

Now, we just need to invoke the method that fits the center and radius properties of a vesicle given a cloud of points. From the command line, this method is called approximateGeometryFromPoints. In the browser dtmslice we can invoke it directly by secondary clicking on one of the points of the model and selecting the option for fitting and plotting.

Menu options after secondary click on a model point
Detailed view on the menu options.

A transparent surface should be visible, allowing you to judge the quality of the fitting quickly sliding the slice up and down.

Depiction of the fitted vesicle surface

From dipole set models

It is also possible to use other models to define vesicle models. This is useful when you have several vesicles in one tomogram, and you don't want to initiate a vesicle model for each one of them. The dipole set model allows creating in a single model several 'dipoles', each one of them having a center, a north and a south. We can use this model to define each vesicle from a dipole: the center property of the dipole can be directly transferred to the vesicle, and the radius of the vesicle can be defined as the distance between the center and north coordinates registered in the dipole.

You can use this code to convert a dipole set into a set of vesicle models. The code assume that the dipoles set model is in the catalogue, and that the resulting models are also saved into the catalogue.

% 
dipoleSetModelFile = '~/first/tomograms/volume_1/models/dipoleset.omd'; % absolute path to your model

% reads the model
ds = dread(dipoleSetModelFile);

% checks the number of individual dipoles
disp(sprintf('Found %d dipoles in the set',length(ds.dipoles)));

% loops on all the dipoles present in the set
for i=1:length(ds.dipoles);
    
    % selects the i-th individual dipole in the set
    dipole = ds.dipoles{i};
       
    % creates a vesicle model   
    v = dmodels.vesicle();

    % prepares a name for each individual dipole
    v.name = sprintf('%srawVesicle_%d',ds.name,i);
   
    % links the created model with the catalogued volume 
    % of the dipole set module
    v.cvolume = ds.cvolume;
    
    % initiates the vesicle model
    v.center = dipole.center;
    v.radius = norm(dipole.center - dipole.north);
    
    % saves the model back into the catalogue
    v.saveInCatalogue();
     
end

Command line

Here come some examples of creation and manipulation of vesicle models from the command line

Creation through user points

The method is called approximateGeometryFromPoints. If a vesicle object v has a set of points stored in the points property (i.e., v.points contains a matrix of Nx3 numbers, each one a 3d coordinate of the vesicle membrane), invoking the order v.approximateGeometryFromPoints will write into v.radius and v.center the codes estimation for radius and center.

Creation through center and radius

This code merely explains how to create an uniform distribution of points on a vesicle model. The central idea is to define the distance between points with the separation property (in pixels), and then invoke the method updateCrop that produces the crop_point coordinates.

a = dmodels.vesicle(); % creates an empty vesicle model
a.radius = 70;
a.center = [40,100,200];
a.separation = 15;
a.updateCrop();  % here it will some time computing the particles.
t=a.grepTable(); % extracts the table
dtplot(t,'profile','oriented_positions'); % scatter plot of images
axis equal; % for a better visualization, same scale an x,y,z
shg

Table plot of positions and particles generated by a vesicle model

Fitting vesicles

Against tomogram data

Dynamo includes some tools for cross correlation based pattern matching. These can be used to fit the center and radius in a roughly picked vesicle. The script that we provide below illustrates how to use the low level vesicle finding utilities in a real life example.

The assumption is that a set of models of approximately picked vesicles (maybe converted from a dipole set) has been defined in the tomogram with index 1 of a catalogue located in the folder '../catalogues/ctfCorrected with relation to the current directory. The script extracts the models whose name follows a pattern as a cell array and then loops on them, fitting them to a simulated vesicle.

The fitting procedure for each vesicle works by first extracting from the tomogram a cube around the vesicle (with extractEnclosingVolume). Then, the low-level fitting tool is called dpkfw.vesicle.ccGlobal.locateRadius, and works by testing a set of predefined radiuses. The radius that yields the largest cross correlation is selected. For each radius, the function computes a global cross correlation between the cropped enclosing volume, thus fitting also the center of the vesicle.

Note that the code below does not overwrite the original models, but manually created a fitted version of each one, which is then saved into the original catalogue.

% extracts model files in a catalogued volume of interest.
files = dynamo_catalogue_models('../catalogues/ctfCorrected','i',1,'nameContains','rawVesicle');

disp(sprintf('Found %d raw vesicle files in the set',length(files)));

% loops on all the dipoles present in the set
for i=1:length(files);
    
    %
    disp(sprintf('Fitting vesicle %d in model file %s',i,files{i}));
    
    % converts the file into  a model object
    raw = dread(files{i});
    
    % prepares another vesicle to contain a fitted version of the original
    % one.
    fitted = copyof(raw);
    % this vesicle model is assigend to the same catalogued volume as the
    % original model 'raw'
    % It also inherits its center and radius properties.
    
    fitted.name = ['fittedVesicle_',num2str(i)];
    
    
    % extracts the cube around the estimated center of the vesicle
    % As a buffer, we add 30 pixels to the estimated radius
    [dataCube,outputEnclosing] = fitted.extractEnclosingVolume('radius',fitted.radius+30);


    if isempty(dataCube)
         % if the volume went out of scope, we desist on analyzing this vesicle
        disp(sprintf('Could not extract data from vesicle %d',i));
        continue;
    end
    
    % pixels of the membrane of the vesicle template that
    membraneThickness = 10;
    
    % set of radii to be tested
    listRadius = fitted.radius-30:  fitted.radius+30;
    
    % It's normally convenient to introduce a binning factor
    % All magnitudes are rescaled automatically, so that all data
    % is expressed in unbinned tompgrams
    binningFactor = 2;
    
    % operates the fitting
    centering  = dpkfw.vesicle.ccGlobal.locateRadius(dataCube,listRadius,....
        'bin',binningFactor,'membraneThickness',membraneThickness,'bar',true);
    
    
    % puts the result of the cross correlation back into the vesicle model
    fitted.radius = centering.r;
    
    % the output of the extraction operation includes an utility to
    % transform coordinates between the extracted enclosing volume and the 
    % original tomogram
    fitted.center = outputEnclosing.dVolume.coords.local2real(centering.center);
    
    % puts the model contanining a fitted vesicle inside the catalogue
    fitted.saveInCatalogue();
    
end