ImagicProtMSA

Gregory Sharov edited this page Jun 15, 2017 · 7 revisions

Scipion Logo


The following two protocols provide wrappers around Multivariate Statistical Analysis (MSA) module of IMAGIC-4D software suite. IMAGIC is a licensed software, not distributed within Scipion and has to be installed by user.

Installation

You will need to use 1.1 version of Scipion to be able to run these protocols. Don’t forget to set IMAGIC_HOME in scipion.conf file to the folder where the IMAGIC is installed (it is the same as IMAGIC_ROOT variable in your shell environment). If you want to use MPI-based parallel job execution, make sure you have openmpi directory inside IMAGIC installation folder. To check the installation, simply run the following Scipion test:

scipion test tests.em.workflows.test_workflow_imagicMSA.TestImagicWorkflow

Supported versions

Since almost with every version of IMAGIC software user interaction with IMAGIC programs is changed, we came up with a way to provide multiple version support. In pyworkflow/em/packages/imagic/scripts directory you have a folder for each corresponding version, that contains batch scripts similar to those used by IMAGIC. This way one can create a similar script specific to a certain version. At this moment, versions 110308 (Mar 2011) and 160418 (Apr 2016) are supported. If you experience any problems or need help with adapting scripts for your IMAGIC version, do not hesitate to Contact us.

imagic - msa

Multivariate Statistical Analysis (MSA) is a powerful technique that allows to identify largest variations in a big data set. It was originally introduced to discriminate between various classes of molecular projections prior to averaging. In the MSA approach, aligned molecular images are submitted to correspondence analysis (CA), that determines the main (orthogonal) directions of inter-image variance and calculates the image coordinates in a system spanned by these newly determined axes. Since this new coordinate system is adapted to the general behavior of the image data, a large reduction in the total amount of data can be obtained: for example, instead of 64x64=4096 density values (pixels) per image, each image is now characterized by the first eight factorial-axis coordinates at the most! With this large data reduction, the classification of the images becomes much simpler.

To launch MSA protocol, you have to provide an aligned (at least, centered) SetOfParticles, number of factors (eigenvectors), maximum number of iterations for algorithm to converge and a mask if you want to analyze variance withing specific area of your particles (fig. 1). Usually 20-25 factors and similar number of iterations are enough even for large data sets.

01.ImagicMsa
Figure 1. GUI input form of the imagic - msa protocol


If you want to play with advanced parameters, select Advanced expert level and look at the Help message for any particular option.

02.ImagicMsaAdvanced
Figure 2. Advanced protocol parameters


This protocol does not generate any results except eigenimages. Eigenimages represent eigenvectors in the image space and account for major density variations in the data set (fig. 3). The very first eigenimage is a total sum of all particles. The following eigenimages show data set variance in a decreasing order. Last eigenimages are usually very noisy and can be discarded from further analysis.

03.Eigenimages
Figure 3. Displaying results of MSA protocol

imagic - msa-classify

After MSA analysis you can use a subset of eigenimages for clustering original images (that will be reconstructed from a linear combination of selected eigenvectors) into groups. IMAGIC MSA module implements hierarchical ascendant classification (HAC) that tries to merge images into clusters by minimizing intra-class variance and maximizing inter-class variance between different clusters.

The msa-classify protocol requires the SetOfParticles from the previous run of msa, a number of factors to use for analysis and a number of classes. At this moment only first N eigenimages can be chosen for MSA-based classification. In the future versions of the protocol it will be possible to select eigenimages independently and also assign weighting coefficients for more advanced image analysis.

04.ImagicMsaClassify
Figure 4. GUI input form of the imagic - msa classify protocol

As always, if you want to play with advanced parameters, select Advanced expert level and look at the Help message for any particular option.

05.ImagicMsaClassifyAdvanced
Figure 5. Advanced protocol parameters

In the end you will obtain 2D classes that will most likely display what kind of heterogeneity you have in your data set.

06.Classes 2D
Figure 6. Output 2D classes


References

  • [[[1]]] M van Heel and W Keegstra (1981). IMAGIC: A fast, flexible and friendly image analysis software system. Ultramicroscopy 7: 113-130.

  • [[[2]]] M van Heel, G Harauz, EV Orlova, R Schmidt and M Schatz (1996). A new generation of the IMAGIC image processing system. J. Struct. Biol. 116: 17-24.

  • [[[3]]] M van Heel, R Portugal, A Rohou, C Linnemayr, C Bebeacua, R Schmidt, T Grant and M Schatz (2012). Four-Dimensional Cryo Electron Microscopy at Quasi Atomic Resolution: "IMAGIC 4D”. International Tables for Crystallography, vol. F, ch. 19.9: 624-628.

  • [[[4]]] M van Heel (1984). Multivariate statistical classification of noisy images (randomly oriented biological macromolecules). Ultramicroscopy 13(1-2): 165-183.

  • [[[5]]] Lisa Borland and Marin van Heel (1990). Classification of image data in conjugate representation spaces. Journal of the Optical Society of America A 7(4): 601-610.

Clone this wiki locally
You can’t perform that action at this time.
You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session.
Press h to open a hovercard with more details.