Identification of Proteins from Mass Spectrometry Data: A Statistical Approach

描述（由申请人提供）： 国家癌症研究所（NCI-CPTC）的癌症临床蛋白质组学技术倡议得出结论，质谱（MS）是监测生物液体，细胞和组织的主要平台之一。尽管在获得精确的质谱仪方面已经取得了相当大的进展，但仍然需要开发先进的定量算法和软件来分析质谱数据，从而产生可再现的结果。此外，虽然识别单个癌蛋白是必要的，但阐明这些蛋白的相互作用也是相关的。了解蛋白质-蛋白质相互作用网络（PIN）的实验举措虽然有价值，但在过去并没有产生一致和可重复的结果。因此，仍然需要通过计算来预测PIN。酵母双杂交（Y2 H）数据和亲和纯化质谱（AP-MS）数据有许多计算预测程序。然而，这些实验是昂贵的。在这里，我们提出了一个完全不同的方法，建立一个大规模的蛋白质共现相互作用网络（PCN）从一个基本的碎片肽MS/MS数据。这个PCN背后的逻辑是，如果两种蛋白质以高概率在样品中共存，那么它们相互作用的机会就更高。因此，这可以提供进行昂贵的诱饵-猎物实验，如Y2 H和AP-MS的蛋白质-蛋白质相互作用的实验验证的计算准则。此外，我们还比较了 PCN用于患病和正常样品，以确定作为癌症治疗的潜在生物标志物或药物靶标的最重要的蛋白质组。我们的目标是在这个建议是三重：1）提出一种新的分层贝叶斯统计方法，以确定从宫颈癌和乳腺癌患者的液体和组织样本的MS/MS光谱的蛋白质。2)使用扩展的分层贝叶斯模型构建癌症蛋白质组的PCN;此外，解开PCN的关键拓扑特征，如枢纽，模块，子网和瓶颈。3)使用统计推断区分患病样本和对照样本之间PCN的关键特征的PCN总体结构。我们预计，该项目的完成将提高我们对宫颈癌和乳腺癌的理解，并有助于开发新的癌症药物。此外，它还将培养21世纪计算生物医学研究领域的博士生。

DESCRIPTION (provided by applicant): Clinical Proteomic Technologies for Cancer initiative of the National Cancer Institute (NCI-CPTC) concluded that mass spectrometry (MS) is one of the main platforms for monitoring biological fluids, cells and tissues. Although, there has been considerable progress in acquiring accurate mass spectrometers, there is still a need for development of advanced quantitative algorithms and software to analyze mass spectrometry data leading to reproducible results. Additionally, although it is essential to identify individual onco-proteins, it is also pertinent t elucidate the interplay of these proteins. Experimental initiatives to understand the protein-protein interaction network (PIN), although valuable, did not produce consistent and reproducible results in the past. Hence, there continues to be a need to predict the PIN computationally. There are numerous computational prediction procedures for Yeast Two Hybrid (Y2H) data and Affinity Purification Mass Spectrometry (AP-MS) data. However, these experiments are expensive. Here, we propose a completely different approach of building a large scale protein co-occurrence interaction network (PCN) from a basic fragmented peptide MS/MS data. The logic behind this PCN is that if two proteins co-occur in a sample with high probability then their chance of interaction is higher. Hence this could provide computational guidelines of conducting expensive bait-prey experiments like Y2H and AP-MS for experimental validation of a protein-protein interaction. In addition, we compare the differential nature of the PCN for the diseased and the normal samples to determine the most important protein groups that are potential biomarkers or drug targets for cancer treatments. Our goals in this proposal are three-fold: 1) Propose a novel hierarchical Bayesian statistical methodology to identify proteins from MS/MS spectra obtained from fluids and tissue samples of cervical and breast cancer patients. 2) Use an expanded hierarchical Bayesian model to construct the PCN of the cancer proteome; additionally, unravel the key topological features of PCNs such as hubs, modules, sub-networks and bottlenecks. 3) Use statistical inference to differentiate the overall structure of the PCNs for the key features of the PCNs between the diseased and control samples. We anticipate that completeness of the project will enhance our understanding of cervical and breast cancer and aid in developing novel cancer drugs. Additionally, it will train doctoral students in the field of computational biomedical research of the twenty-first century.