Institute of Biomaterials and Biomedical Engineering, University of Toronto, Toronto, ON M5S 3G9, Canada; Ted Rogers Centre for Heart Research, University of Toronto, Toronto, ON M5G 1M1, Canada.
Institute of Biomaterials and Biomedical Engineering, University of Toronto, Toronto, ON M5S 3G9, Canada; Ted Rogers Centre for Heart Research, University of Toronto, Toronto, ON M5G 1M1, Canada; Department of Cell and Systems Biology, University of Toronto, Toronto, ON M5S 3G5, Canada; Developmental and Stem Cell Biology Program, The Hospital for Sick Children, Toronto, ON M5G 1X8, Canada.
Biochim Biophys Acta Proteins Proteom. 2017 Nov;1865(11 Pt B):1719-1727. doi: 10.1016/j.bbapap.2017.09.013. Epub 2017 Sep 30.
The development of new microscopy techniques for super-resolved, long-term monitoring of cellular and subcellular dynamics in living organisms is revealing new fundamental aspects of tissue development and repair. However, new microscopy approaches present several challenges. In addition to unprecedented requirements for data storage, the analysis of high resolution, time-lapse images is too complex to be done manually. Machine learning techniques are ideally suited for the (semi-)automated analysis of multidimensional image data. In particular, support vector machines (SVMs), have emerged as an efficient method to analyze microscopy images obtained from animals. Here, we discuss the use of SVMs to analyze in vivo microscopy data. We introduce the mathematical framework behind SVMs, and we describe the metrics used by SVMs and other machine learning approaches to classify image data. We discuss the influence of different SVM parameters in the context of an algorithm for cell segmentation and tracking. Finally, we describe how the application of SVMs has been critical to study protein localization in yeast screens, for lineage tracing in C. elegans, or to determine the developmental stage of Drosophila embryos to investigate gene expression dynamics. We propose that SVMs will become central tools in the analysis of the complex image data that novel microscopy modalities have made possible. This article is part of a Special Issue entitled: Biophysics in Canada, edited by Lewis Kay, John Baenziger, Albert Berghuis and Peter Tieleman.
新的显微镜技术的发展，用于超分辨、长期监测活生物体中的细胞和亚细胞动力学，揭示了组织发育和修复的新的基本方面。然而，新的显微镜方法提出了几个挑战。除了对数据存储的前所未有的要求外，高分辨率、时程图像的分析过于复杂，无法手动完成。机器学习技术非常适合于（半自动）分析多维图像数据。特别是支持向量机（SVM）已经成为分析从动物获得的显微镜图像的有效方法。在这里，我们讨论了使用 SVM 来分析体内显微镜数据。我们介绍了 SVM 背后的数学框架，并描述了 SVM 和其他机器学习方法用于分类图像数据的指标。我们讨论了不同 SVM 参数在细胞分割和跟踪算法中的影响。最后，我们描述了 SVM 的应用如何在酵母筛选中的蛋白质定位、秀丽隐杆线虫中的谱系追踪，或确定果蝇胚胎的发育阶段以研究基因表达动力学等方面发挥了关键作用。我们提出，SVM 将成为分析新显微镜模式带来的复杂图像数据的核心工具。本文是题为“加拿大的生物物理学”的特刊的一部分，由 Lewis Kay、John Baenziger、Albert Berghuis 和 Peter Tieleman 编辑。
