Development of an integrated prognostic score using common risk variants and RNA-expression data from primary breast cancers for improved prognostication of outcomes across populations

项目总结/摘要 我的短期目标是将生物信息学的培训与临床肿瘤学的博士后奖学金相结合， 接受过统计遗传学的培训。生殖系数据与体细胞肿瘤发展和临床的相关性 由于生殖细胞和体细胞检测之间的临床分离，结果仍然知之甚少。体细胞 乳腺肿瘤的表达仅在乳腺癌受体阳性的患者中提供临床诊断和预测。 患者这种限制特别影响非洲血统的妇女，她们的激素水平更高， 受体阴性和HER 2阳性疾病。体细胞表达测定也不能准确反映 非洲血统的HIV受体阳性女性的结果。该人群的原发性肿瘤 表现出更具侵略性的分子特征（如更多的同源重组缺陷，更大的 染色体不稳定性和更多的TP 53突变）， 这表明常见的生殖系变异可能会影响体细胞癌症发展的轨迹， 临床结果。我们假设常见的生殖系变异和体细胞肿瘤表达的整合 这些数据将为乳腺癌亚型和人群中的妇女提供预后信息。为了验证这个，我们 将：（1）开发一个综合预后评分，包括常见的生殖系乳腺癌风险 来自乳腺癌患者的变体和RNA表达数据以及用于预测乳腺癌临床预后的测试。 结果。摘要-PrediXcan是一种计算方法，其采用摘要级GWAS和表型， 兴趣作为输入，测试每个基因表达变化如何影响表型，并输出基因水平 关联结果。我们将使用来自最大的乳腺癌风险荟萃分析的汇总级GWAS数据 超过225，000例欧洲血统病例和对照。我们将使用RNA来改进/训练这个分数- 来自癌症基因组图谱（TCGA）中的1，096名患者和来自METABRIC的932名患者的表达数据。 我们将测试TCGA中跟踪的临床结局的预测能力，包括总生存期、疾病特异性 使用考克斯回归分析的生存期和无进展生存期;（2）跨人群的检验转换 综合生殖细胞/体细胞方法预测乳腺癌结果， 非洲血统我们将使用摘要-PrediXcan和摘要级GWAS得出独立评分 来自3个联盟的乳腺癌风险荟萃分析的数据，包括6，522名非洲裔美国人乳腺癌 患者和7,643名对照。我们将使用183名非裔美国TCGA患者测试结果预测， RNA表达和存活数据。我们还将从两个独特的78人队列中对该评分进行试点验证。 尼日利亚患者和35名非洲裔美国患者的RNA-seq和生存数据。当我们进入一个新的时代， 精确肿瘤学，乳腺癌患者无法获得表现良好的预后分层 在不同的人群中。我的目标是学习如何使用现代生物信息学工具， 分层模型可用于亚型和人群，以改善患者护理。

PROJECT SUMMARY/ABSTRACT My short term goal is to integrate training in bioinformatics with a postdoctoral fellowship in clinical oncology and prior training in statistical genetics. Relevance of germline data to somatic tumor development and clinical outcomes remains poorly understood due to clinical separation between germline and somatic testing. Somatic expression from breast tumors offers clinical prognostication and prediction only in hormone-receptor positive patients. This limitation particularly affects women of African ancestry, who experience higher rates of hormone- receptor negative and HER2-positive disease. Somatic expression assays also do not accurately reflect outcomes in hormone-receptor positive women of African ancestry. Primary tumors from this population demonstrate more aggressive molecular features (such as more homologous recombination deficiency, greater chromosomal instability, and more TP53 mutations) relative to those from European-ancestry patients, suggesting that common germline variation may influence the trajectory of somatic cancer development and clinical outcomes. We hypothesize that integration of common germline variation and somatic tumor expression data will yield prognostic information for women across breast cancer subtypes and populations. To test this, we will: (1) Develop an integrated prognostic score incorporating common germline breast cancer risk variants and RNA-expression data from breast cancer patients and test for prediction of clinical outcomes. Summary-PrediXcan is a computational method that takes summary-level GWAS and phenotype of interest as inputs, tests how expression changes in each gene affect phenotype, and outputs gene-level association results. We will use summary-level GWAS data from the largest meta-analysis of breast cancer risk with more than 225,000 European-ancestry cases and controls. We will refine/train this score using RNA- expression data from 1,096 patients in The Cancer Genome Atlas (TCGA) and 932 patients from METABRIC. We will test the predictive power of clinical outcomes tracked in TCGA, including overall survival, disease-specific survival, and progression-free survival using Cox regression analysis; (2) Test translation across populations of an integrated germline/somatic approach to breast cancer outcome prediction using patients of African ancestry. We will derive an independent score using summary-PrediXcan with summary-level GWAS data from a 3-consortia meta-analysis of breast cancer risk that includes 6,522 African-American breast cancer patients and 7,643 controls. We will test outcome prediction using 183 African-American TCGA patients with RNA-expression and survival data. We will also pilot validation of this score from two unique cohorts of 78 Nigerian patients and 35 African-American patients with RNA-seq and survival data. As we enter a new era in precision oncology, patients with breast cancer do not have access to prognostic stratification that performs well in diverse populations. My goal is to learn how to use modern tools of bioinformatics to develop a prognostic stratification model usable across subtypes and populations for improved patient care.