用于纵向组学数据统计分析的GEE-TGDR算法的开发和应用

特征选择算法在纵向组学数据分析中的应用远远落后于它在横向组学数据的数据分析当中的应用。在此，我们旨在开发一种新型的适用于纵向组学数据的特征选择算法：GEE-TGDR。 基于GEE-TGDR结合阈值梯度下降调整法（TGDR）和广义估量方程（GEE）两种成熟稳定的数学模型，我们预期它将继承这两个模型的特色和优势。对此我们除了理论的推算外，更多地是利用模拟数据和实例进行验证。后期我们还会推出一系列配套的R软件包，从而使GEE-TGDR算法的推广应用成为可能。GEE-TGDR在临床研究中，特别是诊断和预后研究中意义重大。它的成功开发及推广应用促进制定“个体化用药”方案的可能，从而优化治疗效果，节约医疗成本，提高病患的存活率和生活质量。

Feature selection algorithm has been commonly used in “omics” data, however, it is uaually limited to cross-sectional analyses. Here, we propose a framework which integrates two mature models; Generalized Estimating Equation (GEE) and Threshold Gradient Descent Regularization (TGDR), into a new algorithm to specifically tackle with feaure selection in a longitudinal “omics” data . ..The framework is referred to as GEE-TGDR. It is expected that GEE-TGDR can inherit the characteristics of both TGDR and GEE, which needs both theoretical justifications and comprehensive evaluation using synthesized and real data. Moreover, after by applying GEE-TGDR to many suitable “omics” data, we will trim the framework to optimize its performance and then issue a series of R BioConductor packages. ..GEE-TGDR is a handy tool in the field of clinical epidemiology, especially for a diagnostic and prognostic study. Moreover, it makes personalized medicine possible, which means more subject-specific effective treatments, less medical expenses, and better surival rate and quality of life for patients.