Longitudinal Studies with Gene Expression Data

描述（由申请人提供）：使用新的现有生物技术，典型的纵向研究跟踪受试者（或一般生物体）样本，可能涉及在一个或多个时间点收集数千维基因表达谱（或一般生物标志物），随时间变化的医疗方法，以及关于临床结果（如生存）的缺失/审查数据。除了涉及基因表达的“审查”和“混淆治疗”等更经典的问题外，此类数据的一个重要问题是，每个基因本身代表重要参数，因此通常需要一次估计数千个参数。特别是，这意味着在估计的参数比独立观测值多的情况下，需要进行统计推断。此外，需要可视化技术和复杂的监督/无监督聚类技术来发现有意义和重要的整体结构。本研究将发展统计半参数方法，用于分析收集基因表达数据的观察性纵向研究中产生的数据。这项工作的一个重要重点将是与主题专家合作，应用所提出的方法来分析纵向研究：1)空气污染对儿童哮喘自然史的因果影响；2)闲暇时间活动对老年人生存和健康的因果影响；3)癌症患者复发/生存与基因表达谱之间的因果关系；4)酵母数据集中的基因表达及其与非编码区的关系。这些方法将使人们有可能了解不同基因的表达（以及编码的蛋白质）如何相互作用，提供对生物化学途径的深入了解，并在基因组水平上提供潜在因果机制的线索。特别是，人们认为基因表达是癌症进展的一个重要指标，了解哪些基因在这一过程中是活跃的，将最终导致更好的诊断和治疗策略。

DESCRIPTION (provided by applicant): With new existing biotechnologies, a typical longitudinal study following up a sample of subjects (or organisms, in general) can involve collection of multi-thousand dimensional gene expression profiles (or biomarkers, in general) at one or more points in time, change of medical treatments over time, and missing/censored data on clinical outcomes such as survival. An important issue with such data -- in addition to the more classical issues "censoring" and "confounding of treatment" involving gene expression is that each gene represents important parameters itself, so that one typically needs to estimate thousands of parameters at once. In particular, this implies that one needs statistical inference in a setting where one estimates many more parameters than one has independent observations. In addition, visualization techniques and sophisticated supervised/unsupervised clustering techniques are required to discover the significant and important overall structures. This research will develop statistical semi parametric methods for the analysis of data arising in observational longitudinal studies collecting gene-expression data. An important focus of this work will be to apply the proposed methods to analyze longitudinal studies in collaboration with subject matter experts on 1) the causal effect of air pollution on the natural history of asthma in children, 2) the causal effect of leisure time activity on survival and health in the elderly population, 3) causal relationships between recurrence/survival and gene expression profiles in cancer patients and 4) gene expression in yeast data sets and its relationship to the non-coding regions. The methods will make it possible to learn how expression of different genes (and hence the encoded proteins) interact, providing insight into biochemical pathways and clues about underlying causal mechanisms at work on the genomic level. In particular, it is believed that gene expression is an important indicator of cancer progression and that an understanding of which genes are active in this process will ultimately lead to better strategies for diagnosis and treatment.