Adaptive Methodology for Functional Biomedical Data

描述(由申请人提供)：越来越多的生物医学研究产生在精细网格上采样的功能数据。这些类型的数据往往是高维和复杂的，具有许多不规则的特征，如峰值和变化点。目前缺乏现有的严格的统计方法来分析这类数据。该研究计划的目标是开发新的贝叶斯方法，为曲线样本的建模和执行推理提供一个统一的框架，该框架足够灵活，可以应用于各种应用，来自各种实验设计，并可以回答广泛的研究问题。1.我们将在基于小波的函数混合模型框架内开发新的方法，该框架包括外围曲线、更广泛的曲线内协方差结构和更高维的函数数据，使其适用于更广泛的函数数据。2.我们将开发基于功能数据(例如蛋白质组图谱)对个体进行分类的方法，使我们能够将多个来源的功能和标量因素的信息结合在一起。我们将开发执行贝叶斯功能假设检验的方法。3.我们将开发自适应方法，将功能预测因子与功能反应联系起来。4.我们将开发自适应函数主成分分析和基于主成分的函数混合模型的方法，这是一个数据驱动的建模框架，在考虑函数数据中可能存在的复杂结构方面非常灵活。5.我们将把这些方法应用于产生功能数据的一些癌症相关研究，包括各种类型的蛋白质组学和基因组学数据。6.我们将开发高效、易于使用、免费提供的代码，以适应本提案中描述的方法。公共卫生相关性：这项研究将开发灵活的新定量方法，可用于回答来自复杂功能数据的广泛研究问题，由于技术创新产生了自动和高维生物测量，生物医学研究中越来越多地遇到这些问题。我们的方法是避免做出不符合数据的限制性简化假设，而是开发能够适应其复杂特征的灵活方法，从而提取它们包含的关于其潜在生物过程的丰富信息。

DESCRIPTION (provided by applicant): An ever-increasing number of biomedical studies yield functional data sampled on a fine grid. These type of data are frequently high dimensional and complex with many irregular features like peaks and change points. There is currently a dearth of existing rigorous statistical methods for analyzing this type of data. The goal of this research program is to develop new Bayesian methodology that provides a unified framework for modeling and performing inference on samples of curves that is flexible enough to apply to a variety of applications, from various experimental designs, and can answer a broad range of research questions. 1. We will develop new methodology within the wavelet-based functional mixed model framework that accommodates outlying curves, a broader class of within- curve covariance structures, and higher dimensional functional data, making it applicable to a broad range of functional data. 2. We will develop methods to classify individuals based on their functional data, e.g. proteomic profiles, in a way that allows us to combine information across functional and scalar factors of multiple sources. We will develop methods to perform Bayesian functional hypothesis testing. 3. We will develop adaptive methods for relating functional predictors to functional responses. 4. We will develop methods for adaptive functional principal components analysis and for principal component-based functional mixed models, which represents a data-driven modeling framework that is extremely flexible in taking into account the complex structure that may be present in the functional data. 5. We will apply the methods to a number of cancer-related studies yielding functional data, including various types of proteomics and genomics data. 6. We will develop efficient, easy-to-use, freely available code to fit the methods described in this proposal. PUBLIC HEALTH RELEVANCE: This research will develop flexible new quantitative methods that can be used to answer a wide range of research questions from complex functional data, which are increasingly encountered in biomedical research as a result of the technological innovations yielding automatic and high dimensional biological measurements. Our approach is to avoid making restrictive simplifying assumptions that do not fit the data, but instead to develop flexible methods that can adapt to their complex features, and thus extract rich information they contain regarding their underlying biological processes.