Methods for high-dimensional data in HIV research

描述（由申请人提供）：我们的研究旨在开发和评估艾滋病研究中高维数据分析的统计方法。现在大量的病毒和宿主特异性生物信息提供了一个令人兴奋的机会，可以根据患者及其感染病毒群的特定特征来定制治疗决策。然而，由于大量潜在的相关参数和它们之间复杂的、未表征的关系，如何最好地使用这些信息创造了一个分析挑战。我们的研究将整合和推进几种分析方法，包括聚类分析、递归划分、混合效应建模、马尔可夫建模和潜在类别建模，最终达到延缓临床疾病和死亡的最佳策略。通过发展新的统计方法，我们将利用病毒基因序列和细胞免疫调节的信息来实现以下具体目标：(1)通过（1a）结合降维技术和马尔可夫模型（1b）扩展潜在转移建模框架，以处理在单个时间点属于多个状态的个体；(2)通过（2a）将基于预测的分类扩展到CD4计数随时间变化，评估细胞免疫调节对CD4计数变化的预测贡献相关数据设置和（2b）扩展潜在类模型以适应状态随时间的变化。我们的方法将广泛应用于艾滋病毒/艾滋病研究的几个领域。拟议的研究将包括将我们的方法应用于两种临床数据设置：(1)在3项Efavirenz临床研究中获得的公开可用的病毒遗传学数据集；(2)目前在一项比较HIV患者结构化治疗中断与持续治疗的临床研究中收集的数据子集。

DESCRIPTION (provided by applicant): Our research aims to develop and evaluate statistical methods for analyzing high-dimensional data in HIV research. The vast array of viral and host specific biological information now available presents an exciting opportunity to tailor treatment decisions to the specific characteristics of patients and their infecting viral populations. However, how best to use this information creates an analytic challenge due to the large number of potentially relevant parameters and the complex, uncharacterized relationships among them. Our research will integrate and advance several analytic methods including cluster analysis, recursive partitioning, mixed effects modeling, Markov modeling and latent class modeling to arrive ultimately at the best strategies for delaying clinical disease and death. Through the development of novel statistical methods, we will draw from information on viral genetic sequences and cellular immune modulation to achieve the following specific aims: (1) To characterize the progression from sensitive to resistant virus over time and the mediating role of treatment exposure through (1a) combining dimension reduction techniques and Markov models and (1b) extending the latent transition modeling framework to handle an individual belonging to multiple states at a single time point and (2) To assess the predictive contribution of cellular immune modulation on changes in CD4 count over time through (2a) extending prediction based classification to the correlated data setting and (2b) extending the latent class model to accommodate changes in state over time. Our methods will apply broadly to several areas of HIV/AIDS research. The proposed research will include the application of our methods to two clinical data settings: (1) a publicly available viral genetics dataset obtained during 3 clinical studies of Efavirenz and (2) a subset of data currently being collected in a clinical study comparing structured treatment interruption to continuous therapy in HIV patients.