Automated analysis of high dimensional flow cytometry data

近年来流式细胞术（FCM）产生的数据量的增加提出了独特的信息学和统计学挑战。人们普遍认识到，这些技术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实验都是通过耗时和主观的一个或两个维度的连续检查来进行视觉分析的。与人工分析相比，我们过去六年的进展表明，计算分析可以是稳健的、可重复的和快速的。我们通过开发算法完成了NSERC DG前一项的短期目标：(1)监督细胞群识别；(2)无监督细胞群体识别中性能最好的机器学习方法；(3)性能最好的生物标志物发现算法；(4)一种无监督的细胞群体鉴定方法可以结合多个试管样本的数据来鉴定细胞群体。我们进行了广泛的KT活动，将这些算法应用于来自全球合作者网络的复杂数据集，以测试他们对化合物作用机制等问题的假设。我们的结果支持当前提案的目标。长期目标：开发一个用户友好的，强大的，免费/开源的计算平台，用于高通量的FCM数据分析，广泛应用于自然科学研究。短期目标：目标1。开发一种强大的方法来匹配样本中常见的细胞群概况，以揭示未标记样本组之间以前未知的关系。目标2。通过实现并行化扩展flowType以支持HPC。通过新的数据汇总统计提高flowType的性能。目标4。将算法应用于合作者的自然科学数据集。短期目标将通过2名博士生的努力，在本科生的支持下完成。我的学术生涯的影响将大大大于社区努力利用我们提出的算法开发比任何我可以单独完成。在这项建议的范围内也是如此。我们开发的一种匹配细胞种群谱的方法将立即应用于国际小鼠表型联盟（International Mouse Phenotyping Consortium），该联盟耗资9亿美元，旨在确定哺乳动物基因组中每个基因的功能。我们与Genentech在开发下一代flowType/RchyOptimyx方面的合作将立即应用于了解新化合物如何影响免疫系统的潜在机制。所有HQP都培养数据科学家的技能，以满足工业界和学术界对这一领域科学家的高需求。

Background The recent increase in the amount of data generated by flow cytometry (FCM) poses unique informatics and statistical challenges. It is widely recognized that one basic challenge for these technologies 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. Progress in research activities from the previous grant (2013+) In contrast to manual analysis, our progress over the last six years has shown computational analysis can be robust, reproducible and rapid. We completed the short term objectives of the previous of the NSERC DG through the development of algorithms for: (1) Supervised cell population identification; (2) The top-performing machine learning approach for unsupervised cell population identification; (3) The top-performing biomarker discovery algorithm; (4) A unsupervised method for cell population identification can incorporate data from multiple tube samples to identify cell populations. We undertook extensive KT activities to apply these algorithms to complex datasets from a worldwide network of collaborators to test their hypothesis on such matters as the mechanism of action of compounds. Our results support the Objectives of the current proposal. Objectives Long term objective: Develop a user-friendly, robust, free/open source computational platform for the high-throughput analysis of FCM data that becomes widely applied for Natural Science research. Short-term objectives: Objective 1. Develop a robust approach to match cell populations profiles in common across samples to uncover previously unknown relationships between groups of unlabelled samples. Objective 2. Extend flowType to support HPC by implementing parallelization Objective 3. Improve the performance of flowType through new data summary statistics. Objective 4. Apply algorithms to collaborator's natural science datasets. Methodology The short-term objectives will be completed through the efforts of 2 PhD students, supported by undergraduates. Impact The impact of my academic career will be significantly larger as a result of community efforts leveraging the algorithms we propose to develop than any I could accomplish solely on my own. This is true also within the context of this proposal. Our development of an approach to match cell population profiles will be immediately applied in the context of the International Mouse Phenotyping Consortium, a $900M effort targeted to the identification of the function of every gene in a mammalian genome. Our work with Genentech in the development of the next iteration of flowType/RchyOptimyx will immediately be applied to understand the mechanisms underlying how new chemical compounds affect the immune system. All HQP develop skills as data scientists, addressing the high demand for scientists in this area in industry and academia.