Galinsky Vitaly L
Illumina, Inc., 9885 Towne Centre Dr., San Diego, CA 92121, USA.
Bioinformatics. 2003 Sep 22;19(14):1832-6. doi: 10.1093/bioinformatics/btg260.
In the first part of this paper the author presented an efficient, robust and completely automated algorithm for spot and block indexing in microarray images with rectangular grids. Although the rectangular grid is currently the most common type of grouping the probes on microarray slides, there is another microarray technology based on bundles of optical fibers where the probes are packed in hexagonal grids. The hexagonal grid provides both advantages and drawbacks over the standard rectangular packing and of course requires adaptation and/or modification of the algorithm of spot indexing presented in the first part of the paper.
In the second part of the paper the author presents a version of the spot indexing algorithm adapted for microarray images with spots packed in hexagonal structures. The algorithm is completely automated, works with hexagonal grids of different types and with different parameters of grid spacing and rotation as well as spot sizes. It can successfully trace the local and global distortions of the grid, including non-orthogonal transformations. Similar to the algorithm from part I, it scales linearly with the grid size, the time complexity is O(M), where M is total number of grid points in hexagonal grid. The algorithm has been tested both on CCD and scanned images with spot expression rates as low as 2%. The processing time of an image with about 50 000 hex grid points was less than a second. For images with high expression rates ( approximately 90%) the registration time is even smaller, around a quarter of a second.
在本文的第一部分，作者提出了一种高效、稳健且完全自动化的算法，用于对具有矩形网格的微阵列图像进行斑点和块索引。尽管矩形网格目前是微阵列载玻片上对探针进行分组的最常见类型，但还有另一种基于光纤束的微阵列技术，其中探针以六边形网格排列。与标准矩形排列相比，六边形网格既有优点也有缺点，当然需要对本文第一部分提出的斑点索引算法进行调整和/或修改。
在本文的第二部分，作者提出了一种适用于具有以六边形结构排列的斑点的微阵列图像的斑点索引算法版本。该算法完全自动化，适用于不同类型的六边形网格，以及具有不同网格间距、旋转参数和斑点大小的情况。它能够成功追踪网格的局部和全局变形，包括非正交变换。与第一部分的算法类似，它随网格大小线性扩展，时间复杂度为O(M)，其中M是六边形网格中的网格点总数。该算法已在CCD图像和扫描图像上进行测试，斑点表达率低至2%。对于具有约50000个六边形网格点的图像，处理时间不到一秒。对于高表达率（约90%）的图像，配准时间甚至更短，约为四分之一秒。
