Analysis and Statistical Validation of Proteomic Datasets

描述(由申请人提供)： 质谱学(MS)数据的计算分析是一个巨大的挑战，特别是当基于MS的蛋白质组学方法被应用于描述复杂的蛋白质混合物，如人血清或组织。我们建议开发一套统计模型和算法，以实现对基于大规模定量串联质谱仪(MS/MS)的人类临床癌症样本蛋白质组数据集的稳健、准确和透明的分析。为此，我们将1)开发新的数据分析方法和算法，用于统计验证使用任何类型的MS仪器、实验样品制备方案和MS/MS数据库搜索软件产生的MS/MS光谱的多肽分配；2)开发一种集成的、基于概率的信息学方法，用于将多肽组装成蛋白质，并用于推断比较样品之间的蛋白质丰度的同一性和变化，从而增加鸟枪式蛋白质组学方法的能力，以识别低分子量和低丰度蛋白质，区分蛋白质亚型，和检测翻译后处理事件3)引入用于评估MS/MS数据质量的多变量度量，并设计用于重新分析未分配的高质量谱的迭代计算策略4)开发统计模型，用于量化从不同研究收集的多肽和蛋白质鉴定的复合数据库中的错误率，从而实现准确的跨实验室比较、数据挖掘和选择临床样本的靶向蛋白质组图谱的候选对象。我们将把这些方法和工具整合到现有的开源数据分析平台Trans-Proteome Pipeline中，并将向蛋白质组界传播新的工具、统计方法和教育材料。拟议的计算研究的最终目标是能够快速和自动地生成具有准确确定的错误率的高质量蛋白质组数据集，从而消除目前困扰蛋白质组学领域的主要技术障碍之一。

DESCRIPTION (provided by applicant): Computational analysis of mass spectrometry (MS) data presents a significant challenge, especially when MS-based proteomic approaches are applied to profile complex protein mixtures such as human serum or tissues. We propose to develop a set of statistical models and algorithms that will enable robust, accurate, and transparent analysis of large-scale quantitative tandem mass-spectrometry (MS/MS) based proteomic datasets from human clinical cancer specimens. To achieve this we will 1) develop novel data analysis methods and algorithms for statistical validation of peptide assignments to MS/MS spectra generated using any type of MS instrumentation, experimental sample preparation protocols, and MS/MS database search software 2) develop an integrated, probability-based informatics approach for assembling peptides into proteins and for inferring the identities and changes in the abundance of proteins between compared samples, thus increasing the power of the shotgun proteomic approach to identify low molecular weight and low abundance proteins, discriminate between protein isoforms, and detect post-translational processing events 3) introduce multivariate metrics for assessing the quality of MS/MS data and design iterative computational strategies for reanalysis of unassigned high quality spectra 4) develop statistical models for quantifying error rates in composite databases of peptide and protein identifications collected from different studies, thus enabling accurate cross-laboratory comparison, data mining, and selection of candidates for targeted proteomic profiling of clinical samples. We will integrate these methods and tools in the existing open source data analysis platform Trans-Proteomic Pipeline, and will disseminate the new tools, statistical methodologies and educational materials to the proteomic community. The ultimate goal of the proposed computational research is to enable fast and automated generation of high quality proteomic dataset with accurately determined error rates, thus removing one of the main technical barriers currently plaguing the field of proteomics.