Collaborative Research: Scalable and Flexible Algorithms to Detect Structural Change in Complex Sequence Data

科学和工程领域的现代技术产生了规模更大、内容更复杂的数据。使用统计和计算方法理解这些数据是重要的，也是具有挑战性的，特别是确定数据中的趋势。例如来自基因组学的大规模序列数据和来自经济学的市场数据。在基因组学中，研究人员通过检查整个基因组上数十万个生物标记物的测量来寻找拷贝数变异(CNV)。在金融工程中，识别和解释股票价格的突变是有用的。在这些例子中，首要目标是从海量序列数据中发现结构变化。在这个合作项目中，研究人员打算开发和研究理论上合理的、实用的灵活和可移植的策略来分析复杂的序列数据，并将所提出的算法应用于实际数据以进行科学发现。研究人员的目标是开发可扩展和灵活的算法，以识别和推断当代高通量数据的结构变化。特别是，他们将致力于：(A)灵活到足以处理非高斯数据和相关数据的快速变化点检测技术；(B)多序列数据联合分析的新统计框架；(C)推断变化点的理论基础，该变化点可以为检测到的变化点指定显著水平并实现对错误发现率(FDR)的控制；(D)将其应用于CNV数据进行科学发现。此外，研究人员计划为建议的方法开发用户友好和公开可用的软件，以便研究人员可以将建议的方法直接应用于他们的研究问题。

Modern technologies in science and engineering generate data that become bigger in size and more complex in content. It is important and challenging to understand such data using statistical and computational methods, especially to identify trends in the data. Examples include large scale sequence data from genomics and market data from economics. In genomics, researchers search for copy number variations (CNVs) by examining hundreds of thousands of measurements of biomarkers along the whole genome. In financial engineering, it is useful to identify and interpret abrupt changes of stock prices. In these examples, a premier goal is to discover structural changes from massive sequence data. In this collaborative project, the investigators intend to develop and study theoretically sound and practically flexible and portable strategies to analyze complex sequence data and apply the proposed algorithms to real data for scientific discovery. The investigators aim to develop scalable and flexible algorithms to identify and infer structural changes in contemporary high-throughput data. In particular, they will work on (a) fast change-point detection techniques which are flexible enough to handle non-Gaussian data and dependent data; (b) new statistical framework for joint analysis of multiple-sequence data; (c) theoretical foundations for inference of change points which can assign significant levels for detected change points and achieves the control of false discovery rate (FDR); (d) applications to CNV data for scientific discoveries. Moreover, the investigators plan to develop user-friendly and publicly accessible software for the proposed methods so that researchers can apply the proposed methodologies directly to their research problems.