Imputing quantitative mass spectrometry proteomics data using non-negative matrix factorization

项目摘要/摘要 阿尔茨海默病(AD)是一种新兴的全球健康威胁，预计到2019年，发病率将翻一番 2050年。AD是一种由畸形蛋白引起的疾病，AD蛋白质组特征的研究没有取得明显的进展 用质谱仪。然而，数据缺失是解释现有fiCan的一个重要障碍 AD质谱学实验。 缺失是指存在于生物样品中但未被质量检测到的多肽或蛋白质 光谱仪由于各种技术因素的影响。这个项目将通过开发机器学习来解决遗漏问题 在定量质谱学数据中输入或估计缺失值的方法。该项目将得到发展 两种独立的填充方法，一种使用非负矩阵分解，另一种使用深度神经网络。 这些计算方法将提高质谱学实验的重现性和统计能力 并将在现有的蛋白质组学实验中实现新的发现。这些归责方法将适用于 几乎任何类型的质谱学实验-串联质谱标、数据依赖采集、数据独立 获取、光谱计数、无标记量子fi阳离子等。这些归属方法将作为轻量级发布， 开放源码和易于使用的软件包，并可能合并到现有的数据处理工作fl操作系统。 我将通过重新分析几个现有AD蛋白质组的数据来证明这些归因方法的实用性 学习。我的归因方法将识别新的差异表达蛋白、共表达模块和AD 这些现有数据集中的生物标志物。我还将分析未发表的数据独立获取(DIA)蛋白质组学 数据来源于AD患者的脑脊液flUID样本。在这里，我将重点介绍区分生物标记物 患者之间基于遗传背景和共病状态。我还将确定患者的生物标志物 无症状AD。 这项提议开发的归罪方法将使独立AD研究人员未来的发现成为可能。 这一建议与NIA的战略方向相一致，试图“识别和理解基因、分子和 阿尔茨海默病发病的细胞机制。

PROJECT SUMMARY/ABSTRACT Alzheimer's disease (AD) represents an emerging global health threat and is a expected to double in prevalence by 2050. AD is a disease of malformed proteins, and significant progress has been made characterizing the AD proteome with mass spectrometery. However, data missingness represents a significant barrier to the interpretation of existing AD mass spectrometry experiments. Missingness refers to peptides or proteins that are present in the biological sample but are not detected by the mass spectrometer due to various technical factors. This project will address missingness by developing machine learning methods for imputing, or estimating, missing values in quantitative mass spectrometry data. The project will develop two separate imputation methods, one using non-negative matrix factorization and the other deep neural networks. These imputation methods will increase the reproducibility and statistical power of mass spectrometry experiments and will enable new discoveries in existing proteomics experiments. These imputation methods will be applicable to virtually any kind of mass spectrometry experiment – tandem mass tag, data dependent acquisition, data independent acquisition, spectral counts, label-free quantification, etc. These imputation methods will be released as lightweight, open-source and easy-to-use software packages and may be incorporated into existing data processing workflows. I will demonstrate the utility of these imputation methods by reanalysing data from several existing AD proteomic studies. My imputation methods will identify novel differentially expressed proteins, co-expression modules and AD biomarkers in these existing datasets. I will also analyze unpublished data-independent acquisition (DIA) proteomics data derived from AD patient cerebrospinal fluid samples. Here I will focus on identifying biomarkers that differentiate between patients based on genetic background and co-morbidity status. I will also identify biomarkers of patients with asymptomatic AD. The imputation methods developed by this proposal will enable future discoveries by independent AD researchers. This proposal aligns with the NIA Strategic Direction seeking to "identify and understand the genetic, molecular and cellular mechanisms underlying the pathogenesis of AD."