Statistical Methods for Analysis of Tumor Heterogeneity in Genetic Epidemiology

项目摘要/摘要 癌症是世界范围内的主要发病率和死亡率负担。虽然已经取得了很大的进展 虽然已经取得了进展，但消除癌症的目标尚未实现。在目前的拨款中，我们开发了 癌症全基因组关联分析的统计方法，并按起源部位研究癌症。 然而，即使在一个部位，癌症也可以在不同的患者中有不同的突变特征。汇集所有癌症 作为一种疾病发生在一个地点的病例可能会错过重要的临床和病因学见解。最近 技术的进步使大量详细描述体细胞突变成为可能。 为研究肿瘤的异质性提供了独特的机会。这场比赛的目标是 更新是为了继续发展我们的统计方法，用于肿瘤异质性的关联分析 具有临床结果，并用于研究潜在的遗传和环境病因。 在分析体细胞突变数据方面存在挑战。首先，体细胞突变可能只存在于 患者的肿瘤细胞亚群，即所谓的肿瘤内异质性。当我们的应用程序专注于 患者之间的肿瘤异质性，因为肿瘤内的异质性也会影响临床结果， 如果不加以考虑，重要的洞察力可能会被遗漏。目标1的目标是发展统计学 评估体细胞突变与肿瘤的相关性时，考虑肿瘤内异质性的方法 临床结果。第二，发现胚系-体细胞突变联系是非常有兴趣的；然而，尽管 由于技术进步，肿瘤研究比以前大得多，发现的力量 这种联系仍然有限，因为适度的遗传效应和解释多个 与测试数以百万计的变种进行比较。目标2的目标是开发新的筛查策略 在测试全基因组与肿瘤异质性的相关性时优先考虑遗传变异。我们将实现 通过使用加权假设检验框架，允许相关的遗传变异和 连续筛查统计。第三，通常情况下，肿瘤块通常只能从 因此，病例子集和肿瘤测序数据仅适用于该子集。同时，广泛的风险 已经为更大规模的研究收集了因素信息。目标3的目标是开发一个强大的 合并来自较大研究的汇总统计信息的有效方法，以确定 遗传和环境危险因素对特定肿瘤特征癌症发生风险的影响 该方法将应用于结直肠癌的遗传学和流行病学研究 (GECCO，PI：Ulrike Peters；首席生物统计学家：Li Hsu)，包括超过125,000例结直肠癌病例 并以GWAS数据和额外的7000个肿瘤测序数据作为对照。因为我们的方法也是 适用于其他癌症研究，我们将在计算效率和用户友好的情况下实现它们 软件包并通过R/CRAN、R/BioConductor或Github向社区传播。

PROJECT SUMMARY/ABSTRACT Cancer is a major morbidity and mortality burden throughout the world. While much progress has been made, the elimination of cancer has not yet been achieved. In the currently funded grant, we have developed statistical methods for genome-wide association analysis of cancer and studied cancer by the site of origin. However, even within a site, cancer can have distinct mutational profiles across patients. Pooling all cancer cases occurring at one site as one disease may miss important clinical and etiological insights. Recently technology advances have made it possible to characterize somatic mutations at great detail in large numbers of tumors, providing a unique opportunity to study tumor heterogeneity. The objective of this competitive renewal is to continue our statistical methods development for association analyses of tumor heterogeneity with clinical outcomes, and for studying the underlying genetic and environmental etiology. There are challenges in analyzing the somatic mutation data. First, somatic mutation may only exist in a subset of tumor cells of a patient, so called intra-tumor heterogeneity. While our application is focused on tumor heterogeneity across patients, because intra-tumor heterogeneity can also impact clinical outcomes, important insight could be missed if it were not accounted for. The goal of Aim 1 is to develop statistical methods to account for intra-tumor heterogeneity when assessing the association of somatic mutations with clinical outcomes. Second, it is of great interest to discover germline-somatic mutation link; however, despite that tumor studies are considerably larger than before due to technology advances, the power for discovering such links remains limited because of moderate genetic effects and the burden of accounting for multiple comparison from testing millions of variants. The goal of Aim 2 is to develop novel screening strategies for prioritizing genetic variants in testing genome-wide association with tumor heterogeneity. We will achieve optimal power by using the weighted hypothesis testing framework, allowing for correlated genetic variants and continuous screening statistics. Third, it is common that tumor blocks can usually only be retrieved from a subset of cases and tumor sequencing data are thus only available for this subset. Meanwhile, extensive risk factor information has already been collected for the larger study. The goal of Aim 3 is to develop a robust and efficient approach to incorporate the summary statistics information from the larger study for characterizing the effects of genetic and environmental risk factors on risk of developing cancer with specific tumor feature. The methods will be applied to the Genetics and Epidemiology of Colorectal Cancer Consortium (GECCO, PI: Ulrike Peters; Lead Biostatistician: Li Hsu), which includes over 125,000 colorectal cancer cases and controls all with GWAS data and additionally 7,000 tumors sequencing data. As our methods are also applicable to other cancer studies, we will implement them in computationally efficient and user-friendly software packages and disseminate them to the community through R/CRAN, R/Bioconductor, or Github.