Automatic Particle Selection for Single Particle Analysis


Single Particle Analysis (SPA) is the problem of creating a three dimensional reconstruction of a molecular structure, e.g. a protein complex, starting from two dimensional images acquired through transmission electron microscopy. An important sub-problem of SPA is particle selection, or "boxing", which consists in localizing and identifying two dimensional projections of the structure in the original micrographs. Particle selection is one of the main practical bottlenecks of SPA, because of the inherent high level of noise in electron microscopy images and the large number of particles (typically above 100.000) that are necessary to obtain high resolution reconstructions. Therefore, automating particle selection has recently become an active research area in biological image analysis.

We have designed an algorithm for automatic particle selection called TextonSVM. Our approach models the appearance of images of molecular structures based on their texture and applies a support vector machine (SVM) classifier to discriminate them from the background and from broken or otherwise unwanted particles. Furthermore, in order to provide an empirical basis for the study of this task, we have developed an experimental framework for quantitative evaluation and comparison of particle selection algorithms in real data.


In order to promote research on automated particle selection for Single Particle Analysis, we are making available the complete resources from our JSB paper below:
  • Original Datasets All the original micrographs used in our experiments.

  • Source Code: A stand-alone implementation of TextonSVM, written in C++ for Linux platforms.

  • Results: Particles detected by TextonSVM, Signature and ground-truth data to perform quantitative evaluation of particle picking algorithms.
[ manual ] [ data ] [ code ] [ github ]


If you use the resources in this page, please cite our paper:

* "Experimental Evaluation of Support Vector Machine-based and Correlation-based Approaches to Automatic Particle Selection".
P. Arbelaez, B.G. Han, D. Typke, J. Lim, R.M. Glaeser, and J. Malik.
Journal of Structural Biology, 2011.
doi:10.1016/j.jsb.2011.05.017    [pdf]

Back to Berkeley Computer Vision page
Questions--> Pablo Arbelaez