Proteogenomic translator for cancer biomarker discovery towards precision medicine

项目总结 我们的PGDAC的目标是提高我们对肿瘤蛋白质基因组学复杂性的理解。朝向 这个目标，我们的第一个目标是应用多重组学和基于网络的系统学习来揭示病因 用CPTAC研究癌症不同表型的分子调控关系 蛋白质基因组学数据。我们将使用为其量身定做的管道从严格的预处理和质量控制开始 基于MS的蛋白质组学数据，用于检测和纠正批次效应、异常值、样本标记错误以及 归因于缺失的值(目标1.1)。然后，我们将使用新的统计工具对≥6种类型的组学进行联合建模 系统地描述DNA改变的功能影响的数据(如DNA突变、CNA和 甲基化)(目标1.2)。这种顺式/反式调节网络将帮助我们阐明蛋白质或途径是如何 活性是由肿瘤细胞的基因组变化决定的。我们还将构建蛋白质/PTM共表达 基于全球、磷酸、糖基和其他PTM蛋白质组学数据的网络(目标1.3)。在施工时 这些网络，我们将使用和创建先进的计算工具来有效地借用信息 文献、可公开获取的开放数据库和转录简介。此外，我们还将研究细胞类型 使用新的多组学去卷积分析从大块组织中提取成分，并识别免疫亚型 具有不同的免疫激活或逃避机制(目标1.4)。此外，我们还将全面执行 通过利用公开可获得的数据来研究激酶和转录因子的活性 从许多监管网络数据库中处理(目标1.5)。所有AIMS 1.2-1.5都将有助于 收集功能相关的蛋白质/PTM组、共表达网络模块、免疫特征以及 以激酶/转铁蛋白活性评分为指标。然后将测试这些功能和功能集与它们之间的关联 疾病表型(目标1.6)。对于目标1中的所有分析任务，我们将派生出 通过泛癌分析研究多种肿瘤类型的共性和差异。我们的第二个目标是 进一步开发方法、软件和基于网络的工具，以优化我们PGDAC的数据分析。我们会 开发新的统计/计算工具；将这些方法作为计算效率和用户- 友好的软件；并建立综合数据分析管道(目标2.1)。我们还计划开发一套 基于网络的服务，用于查询、可视化和解释CPTAC研究的分析结果(目标2.2)。 我们的第三个目标是提名新的基于蛋白质的癌症生物标记物和药物靶标进行进一步研究。 通过靶向蛋白质组学分析。我们将首先将基于机器学习的预测模型应用于特征和 来自目标1的特征集，以识别预测疾病结果、治疗反应和 治疗上不同的疾病亚型(目标3.1)。我们还将查询与疾病相关的基因、蛋白质和PTM 针对函数扰动数据库(如Lincs L1000数据库)的签名，以确定小的优先级 可测试以减缓肿瘤生长或治疗反应的分子和药物(目标3.2)。

PROJECT SUMMARY The goal of our PGDAC is to improve our understanding of the proteogenomic complexity of tumors. Towards this goal, our First Aim is to apply multiomics and network based system learning to reveal causative molecular regulatory relationships contributing to varieties of phenotypes in cancer using CPTAC proteogenomic data. We will start with rigorous preprocessing and quality control using a pipeline tailored to MS-based proteomics data to detect and correct batch effects, outliers, sample labeling errors, as well as to impute missing values (Aim 1.1). We will then utilize novel statistical tools to jointly model ≥6 types of omics data to systematically characterize functional impact of DNA alterations (such as DNA mutations, CNA, and methylations) (Aim 1.2). Such cis-/trans-regulatory networks will help us to elucidate how protein or pathway activities are shaped by genomic alterations in tumor cells. We will also construct protein/PTM co-expression networks based on global-, phospho-, glyco- and other PTM-proteomics data (Aim 1.3). When constructing these networks, we will use and create advanced computational tools to effectively borrow information from literature, publicly available open databases, and transcriptome profiles. Moreover, we will study cell type composition from bulk tissue using novel multi-omics deconvolution analyses, and identify immune subtypes with distinct immune activation or evasion mechanisms (Aim 1.4). Furthermore, we will perform comprehensive investigation of kinase and transcription factor activities by leveraging publicly available data extracted and processed from many regulatory network databases (Aim 1.5). All Aims 1.2-1.5 will contribute to a large collection of functionally related protein/PTM sets, co-expression network modules, immune signatures, as well as kinase/TF activity scores. These features and feature-sets will then be tested for their associations with disease phenotypes (Aim 1.6). For all analysis tasks in Aim 1, we will derive an integrated view of commonalities and differences across multiple tumor types via Pan-Cancer analyses. Our Second Aim is to further develop methods, software, and web-based tools to optimize the data analyses of our PGDAC. We will develop novel statistical/computational tools; implement these methods as computationally efficient and user- friendly software; and construct an integrated data analysis pipeline (Aim 2.1). We also plan to develop a set of web-based services for querying, visualizing, and interpreting analysis results from CPTAC studies (Aim 2.2). Our Third Aim is to nominate novel protein-based cancer biomarkers and drug targets for further investigation by targeted proteomics assays. We will first apply machine-learning-based prediction models on features and feature-sets from Aim 1 to identify protein biomarkers that predict disease outcome, treatment responses, and therapeutically distinct disease subtypes (Aim 3.1). We will also query disease related gene, protein, and PTM signatures against function perturbation databases, such as the LINCS L1000 database, to prioritize small molecules and drugs that could be tested for attenuating tumor growth or treatment response (Aim 3.2).