Methods and Software for High-dimensional Risk Prediction Research

项目摘要 使用人类基因组发现和其他已建立的风险预测因子进行早期疾病预测是一种有效的方法。 迈向精准医疗的重要一步。然而，开发临床上有用的风险预测的任务 模型受到目前证据状况的阻碍，目前已知的风险预测因子不足 来准确预测大多数人类疾病。随着快速发展的高通量技术和不断- 随着成本的降低，在大规模研究中收集不同类型的组学数据变得可行。而 从这些研究中产生的多水平组学数据为进一步改进新的预测方法提供了很大的希望。 现有的模型，组学数据的高维性，人类疾病的异质性病因，以及 各种级别的组学数据之间的复杂相互关系带来了巨大的分析挑战。新 方法和软件是非常需要的，以解决这些挑战，并促进正在进行的和未来的高， 维度风险预测研究因此，本应用程序的目标是完成 使用组学数据进行高维风险预测研究的随机场（RF）框架和软件，以及 然后将这个框架应用于阿尔茨海默病（AD）。所提出的研究将集成一个核函数 和一个空间自适应套索到RF，使其适用于高维数据与大量的 预测器此外，新框架能够利用族设计来解决几个重要问题 (e.g.,遗传异质性）预测复杂疾病，并将采用交叉扩散过程， 整合来自不同层次omic数据的信息。根据初步的模拟结果，我们的中央 假设所提出的框架比现有的框架获得更准确和鲁棒的性能 方法.这一项目的成功完成应能解决大规模杀伤性武器所面临的分析挑战。 大量的omic数据，并推进高维风险的方法和软件开发 一般的预测。将新方法和软件应用于大规模AD数据集也可以 导致新的AD风险预测模型，可以进一步复制和研究，通过合作 research.

Project Summary The use of human genome discoveries and other established risk predictors for early disease prediction is an essential step towards precision medicine. However, the task of developing clinically useful risk prediction models is hampered by the present state of evidence, in which currently known risk predictors are insufficient for accurately predicting most human diseases. With rapidly evolving high-throughput technologies and ever- decreasing costs, it becomes feasible to collect diverse types of omic data in large-scale studies. While the multi-level omic data generated from these studies hold great promise for novel predictors to further improve existing models, the high-dimensionality of omic data, the heterogeneous etiology of human diseases, and the complex inter-relationships among various levels of omic data bring tremendous analytic challenges. New methods and software are in great need to address these challenges, and to facilitate ongoing and future high- dimensional risk prediction research. The goal of this application is thus to complete the development of a random field (RF) framework and software for high-dimensional risk prediction research using omic data, and then apply the framework to Alzheimer's disease (AD). The proposed research will integrate a kernel function and a spatial adaptive lasso into RF, making it applicable for high-dimensional data with a large number of predictors. Moreover, the new framework is able to utilize the family design to address several important issues (e.g., genetic heterogeneity) in predicting complex diseases, and will adopt a cross-diffusion process to integrate information from different levels of omic data. Based on preliminary simulation results, our central hypothesis is that the proposed framework attains a more accurate and robust performance than existing methods. The successful completion of this project should address analytical challenges faced by massive amounts of omic data, and advance the methodology and software development for high-dimensional risk prediction in general. The application of the new methods and software to large-scale AD datasets could also lead to novel AD risk prediction models that could be further replicated and investigated through collaborative research.