Bayesian Inference for Tumor Heterogeneity with Next-Generation Sequencing Data

 描述（由申请人提供）：为了实现广泛影响癌症诊断和预后的目标，我们提出了使用下一代测序（NGS）数据推断肿瘤异质性（TH）的特征分配模型。基于非参数贝叶斯统计中的印度自助过程（IBP），我们提出了在核苷酸水平上对肿瘤样本中未观察到的亚克隆进行后验推断。亚克隆由独特的DNA序列和拷贝数标记，反映了克隆扩增和肿瘤发生期间发生的变化。我们还将开发有效的计算方法来分析NGS实验产生的大量数据，为使用所提出的方法的实际应用铺平道路。在目标1和2中，我们将重点关注NGS数据中噪声的统计模型开发，并建立可扩展的计算。我们将推广经典的IBP模型，以适应分类和相关的随机矩阵，从而产生cIBP和dIBP模型。在目标3中，我们提出了一个基于TH的个性化癌症治疗临床试验。该试验的一个独特之处是将基于TH的适应性治疗策略与忽略TH的标准固定治疗策略进行了比较。我们打算开发创新和高效的贝叶斯计算方法，使用内部和实验室可用的基因组学数据应用所提出的方法，并通过我们的在线门户网站www.compgenome.org传播所有开发的工具（目标4）。拟议的研究将促进统计方法的进步，并促进新的贝叶斯非参数模型类的发展。此外，随着这种类型的统计学进步，关于肿瘤异质性的重要问题将得到解决。

 DESCRIPTION (provided by applicant): To achieve the goal of broadly impacting cancer diagnosis and prognosis, we propose feature allocation models for the inference of tumor heterogeneity (TH) using next-generation sequencing (NGS) data. Building upon the Indian buffet process (IBP) in nonparametrics Bayesian statistics, we propose posterior inference on unobserved subclones in a tumor sample at the nucleotide level. The subclones are marked by distinctive DNA sequences and copy numbers, reflecting the variations that occur during clonal expansion and tumorgenesis. We will also develop efficient computational approaches for analyzing extensive data generated from NGS experiments, paving ways for real-life applications using the proposed methods. In Aims 1 and 2, we will focus on statistical model development accounting for noises in the NGS data and set up a scalable computation. We will generalize the classical IBP model to accommodate both categorical and dependent random matrices, giving rise to the cIBP and dIBP models. In Aim 3, we propose a TH-based clinical trial for personalized cancer treatment. A unique feature of the trial is its comparison of the adaptive treatment strategies based on TH to a standard, fixed treatment strategy that ignores TH. We intend to develop innovative and efficient Bayesian computational approaches, apply the proposed methods using in-house and publically available genomics data, and disseminate all of the developed tools through our online portal at www.compgenome.org (Aim 4). The proposed research will promote advancement in statistical methodology and foster development of new classes of Bayesian nonparametrics models. Further, with this type of statistical advancement, important questions on tumor heterogeneity will be addressed.