Biocomputing Unit, National Center of Biotechnology (CSIC), Department of Computer Science, University Autonoma de Madrid, Campus Universidad Autonoma s/n, 28049 Cantoblanco, Madrid, Spain, Department Applied Mathematics, Tel Aviv University, Ramat Aviv, Tel Aviv 69978 Israel and Bioengineering Lab, Escuela Politecnica Superior, University San Pablo CEU, 28668 Boadilla del Monte, Madrid, Spain.
Bioinformatics. 2013 Oct 1;29(19):2460-8. doi: 10.1093/bioinformatics/btt429. Epub 2013 Aug 19.
Structural information of macromolecular complexes provides key insights into the way they carry out their biological functions. Achieving high-resolution structural details with electron microscopy requires the identification of a large number (up to hundreds of thousands) of single particles from electron micrographs, which is a laborious task if it has to be manually done and constitutes a hurdle towards high-throughput. Automatic particle selection in micrographs is far from being settled and new and more robust algorithms are required to reduce the number of false positives and false negatives.
In this article, we introduce an automatic particle picker that learns from the user the kind of particles he is interested in. Particle candidates are quickly and robustly classified as particles or non-particles. A number of new discriminative shape-related features as well as some statistical description of the image grey intensities are used to train two support vector machine classifiers. Experimental results demonstrate that the proposed method: (i) has a considerably low computational complexity and (ii) provides results better or comparable with previously reported methods at a fraction of their computing time.
The algorithm is fully implemented in the open-source Xmipp package and downloadable from http://xmipp.cnb.csic.es.
大分子复合物的结构信息为其执行生物功能的方式提供了关键的见解。要通过电子显微镜获得高分辨率的结构细节，需要从电子显微镜图像中识别出大量（多达数十万）的单个粒子，如果必须手动完成，这将是一项费力的任务，并且是高通量的障碍。在显微照片中自动选择粒子还远未解决，需要新的和更强大的算法来减少假阳性和假阴性的数量。
在本文中，我们引入了一种自动粒子选择器，它可以从用户那里学习他感兴趣的粒子类型。粒子候选者可以快速而稳健地分类为粒子或非粒子。使用了一些新的有区别的形状相关特征以及图像灰度强度的一些统计描述来训练两个支持向量机分类器。实验结果表明，所提出的方法：（i）具有相当低的计算复杂度，（ii）以其计算时间的一小部分提供了优于或可与先前报道的方法相媲美的结果。
该算法已在开源 Xmipp 软件包中完全实现，并可从 http://xmipp.cnb.csic.es 下载。
