Heterogeneous Cancer Progression from Microarray Data

描述（由申请人提供）：来自微阵列数据的异质性癌症进展这类统称为癌症的疾病原则上可以由无限数量的突变组合产生。尽管如此，很明显，大多数癌症可以分为几个常见的“亚型”，每个亚型都有一个共同的特点，即对细胞生长的控制失效。通过识别这些常见的亚型和产生这些亚型的特定基因异常序列，我们可以识别对不同治疗反应的患者亚群，找到可能成为新抗癌药物有用靶点的基因，并开发诊断测试来更好地预测患者的结果，并建议哪些药物对哪些患者有益。由于不同的癌症亚型具有过度活跃或过度不活跃基因的特征模式，因此通过检查肿瘤内的基因表达已经取得了很大进展。然而，试图解释这些表达数据是一个难题，复杂的计算机模型已被证明是无价的。一类计算机模型——系统发育（进化树）模型——为解释肿瘤内不同细胞类型进化的可能途径提供了一种强有力的方法。这种系统发育方法有两种重要的变体：一种使用从基因表达微阵列收集的数据，这种数据可以分析大型肿瘤样本中平均数千个基因，另一种使用从细胞分析研究收集的数据，这种数据可以分析从肿瘤中分离的单个细胞中的少量基因。每一种方法都有各自的优点，前者可以更全面地了解整体基因活动，而后者可以通过识别在单个肿瘤中同时出现的细胞类型，为肿瘤进化提供有价值的线索。拟议的工作将为这些问题开发新的计算机模型，以便开发一种具有两种方法优点的单一方法。这项工作将首先开发方法，通过在患者群体中取样的肿瘤的大量微阵列测量来推断常见细胞类型的存在。然后，它将建立在先前的方法上，推断这些肿瘤状态之间的进化相似性。最后，它将使细胞肿瘤系统发育方法适应从这些微阵列状态推断进化序列的问题。结果将是一种统一的方法来推断单个细胞状态之间的进化，就像在细胞分析研究中一样，但在数千个基因上进行分析，就像在微阵列研究中一样。统一的方法将在乳腺癌数据上进行验证，微阵列和细胞术测量都是可用的，并应用于发现乳腺癌人群的共同进展途径。该研究有望揭示现有方法无法发现的乳腺癌进展的不同阶段，有助于确定新的患者亚群、药物靶点和诊断测试。待开发的方法可能具有更广泛的适用性，一般实体肿瘤进展和分析混合样本中细胞分化的相关问题。

DESCRIPTION (provided by applicant): Heterogeneous Cancer Progression from Microarray Data the class of diseases collectively known as cancer could in principle be produced by a limitless number of combinations of mutations. Nonetheless, it has become apparent that most cancers can be grouped into a few common "sub-types," each characterized by a common way in which the controls on cell growth become disabled. By identifying these common sub-types and the particular sequences of genetic abnormalities that produce them, we can identify patient sub-populations who may respond to different treatments than the general population, find genes that may be useful targets for new anti-cancer drugs, and develop diagnostic tests to better predict patient outcomes and suggest which drugs will benefit which patients. Great progress has been made by examining gene expression within tumors, as different cancer sub-types have characteristic patterns of overly active or overly inactive genes. Trying to interpret these expression data is, however, a difficult problem for which sophisticated computer models have proven invaluable. One class of computer models - phylogenetic (evolutionary tree) models - has provided a powerful method for interpreting likely pathways by which different cell types evolve within tumors. There are two important variants of this phylogenetic approach: one using data gathered from gene expression microarrays, which assay thousands of genes averaged over large tumor samples, and another using data gathered from cytometric studies, which assay small numbers of genes in individual cells isolated from tumors. Each has advantages, the former in allowing a far more complete picture of overall gene activity and the latter in providing valuable clues about tumor evolution by identifying which cell types co-occur in individual tumors. The proposed work will develop new computer models for these problems in order to develop a single approach with the advantages of both methods. The work will first develop approaches to infer the existence of common cell types from bulk microarray measurements of tumors sampled across patient populations. It will then build on prior methods to infer evolutionary similarity between these tumor states. It will, finally, adapt methods for cytometric tumor phylogenetics to the problem of inferring evolutionary sequences from these microarray states. The result will be a unified approach for inferring evolution among individual cell states, as in a cytometric study, but assayed on thousands of genes, as in a microarray study. The unified approach will be validated on breast cancer data, for which both microarray and cytometric measurements are available, and applied to the discovery of common progression pathways in breast cancer populations. The study can be expected to uncover distinct stages in the breast cancer progression that would not be apparent by existing methods, aiding in the identification of new patient sub-populations, drug targets, and diagnostic tests. The methods to be developed are likely to have broader applicability to solid tumor progression in general and to related problems of analyzing cell differentiation in mixed samples.