insideOutside: an accessible algorithm for classifying interior and exterior points, with applications in embryology.
insideOutside: an accessible algorithm for classifying interior and exterior points, with applications in embryology.
复制标题
作者：
胚胎学的一个关键方面是将单个细胞的位置与胚胎的更广泛的几何形状联系起来。一个经典的例子是小鼠胚胎的第一个细胞命运决定，其中内部细胞成为内细胞团，外部细胞成为滋养外胚层。荧光标记，成像和定量的组织特异性蛋白质已经推进了我们对这个动态过程的理解。然而，出现了这些标记不可用或不可靠的情况，我们只剩下细胞的空间位置。因此，需要一种使用空间信息对胚胎的内部和外部细胞进行分类的简单、稳健的方法。在这里，我们描述了一个简单的数学框架和一种无监督的机器学习方法，称为insideOutside，用于对三维点云的内部和外部点进行分类，这是早期小鼠胚胎内成像细胞的常见输出。我们对其他已发表的方法进行基准测试，以证明它在对植入前小鼠胚胎的细胞核进行分类时具有更高的准确性，并且在局部表面凹陷的情况下具有更高的准确性。我们已经免费提供了该方法的MATLAB和Python实现。这种方法应该证明是有用的胚胎学，与更广泛的应用，以类似的数据在生命科学中出现。总结：出于对仅基于空间位置的小鼠胚胎细胞进行分类的需要，我们建立了一个理论框架，开发了一个易于部署的软件包，并进行基准测试。
A crucial aspect of embryology is relating the position of individual cells to the broader geometry of the embryo. A classic example of this is the first cell-fate decision of the mouse embryo, where interior cells become inner cell mass and exterior cells become trophectoderm. Fluorescent labelling, imaging, and quantification of tissue-specific proteins have advanced our understanding of this dynamic process. However, instances arise where these markers are either not available, or not reliable, and we are left only with the cells’ spatial locations. Therefore, a simple, robust method for classifying interior and exterior cells of an embryo using spatial information is required. Here, we describe a simple mathematical framework and an unsupervised machine learning approach, termed insideOutside, for classifying interior and exterior points of a three-dimensional point-cloud, a common output from imaged cells within the early mouse embryo. We benchmark our method against other published methods to demonstrate that it yields greater accuracy in classification of nuclei from the pre-implantation mouse embryos and greater accuracy when challenged with local surface concavities. We have made MATLAB and Python implementations of the method freely available. This method should prove useful for embryology, with broader applications to similar data arising in the life sciences. Summary: Motivated by the need to classify cells of mouse embryos based on spatial position alone, we establish a theoretical framework, develop an easy-to-deploy package, and perform benchmarking.