Automated analysis of high dimensional flow cytometry data

流式细胞术（FCM）产生的数据量的最近增加带来了独特的信息学和统计学挑战，并且人们广泛认识到，FCM的一个基本挑战是简化数据和统计信息的提取。传统上，大多数FCM实验都是通过一次对一个或两个维度进行耗时和主观的连续检查，对多达17个维度的数据进行视觉分析。虽然商业上可用的分析工具旨在促进这一过程，这种可视化程序必然会忽略数据的更高维度。此外，视觉分析的主观性是研究自动化和可重复性的主要障碍。与2D视觉分析相比，多变量统计分析可以同时挖掘全套标记。* 在这项研究计划中，我们将开发新的和扩展现有的统计工具，以帮助分析高通量，高维流式细胞术数据。我们将继续我们的工具和算法开发，重点是FCM数据的半监督聚类方法和多个样本等分试样的集成分析，这两个领域尚未在社区内得到充分解决。从长远来看，我们将开发新的方法来自动识别样本组之间的显著差异。 我们将与基础研究人员合作，然后将这些方法应用于几个数据集，通过与生物学家的互动来改进我们的方法，并回答几个生物学问题。一个重要的合作将是与国际小鼠表型协会，这是敲除每个基因，试图了解其功能。这些方法的发展将提高结果的鲁棒性超过人工分析的所有领域，这一技术的应用。这将打开使用FCM作为非假设驱动技术的可能性，使发现不可预见的关系和发现变得合理，否则将无法识别。

The recent increase in the amount of data generated by flow cytometry (FCM) poses unique informatics and statistical challenges, and it is widely recognized that one basic challenge for FCM is to simplify the extraction of data and statistical information. Traditionally, the majority of FCM experiments have been analyzed visually, through time-consuming and subjective serial inspection of one or two dimensions at a time, of up to 17 dimensional data. While commercially available analysis tools are designed to facilitate this process, such visual procedures necessarily neglect the higher dimensionality of the data. Furthermore, the subjective character of visual analysis is a major obstacle to the automation and reproducibility of research. In contrast to 2D visual analysis, a multivariate statistical analysis can mine the full set of markers simultaneously. ****In this research program we will develop novel and extend existing statistical tools to assist in the analysis of high throughput, high dimensional flow cytometry data. We will continue our tool and algorithm development with a focus on semi-supervised clustering methods for FCM data and integrating analysis across multiple sample aliquots, two areas that have not yet been adequately addressed within the community. In the longer term we will develop new approaches for automatically identifying what are the significant differences between sample groups. We will work with basic researchers to then apply these methods to several datasets to both improve our methods through interactions with biologists, and to answer several biological questions. An important collaboration will be with the International Mouse Phenotyping Consortium, which is knocking out each gene in an attempt to understand its function. The development of these methodologies will improve the robustness of results over the manual analysis for all the areas where this technology is employed. This will open the possibility to use FCM as a non-hypothesis driven technology, making plausible the discovery of unforeseen relationships and discoveries that would not have been otherwise identified.****