Reliable Identification of Significant Differences in Differential Hydrogen Exchange-Mass Spectrometry Measurements Using a Hybrid Significance Testing Approach
Reliable Identification of Significant Differences in Differential Hydrogen Exchange-Mass Spectrometry Measurements Using a Hybrid Significance Testing Approach
复制标题
DOI：
10.1021/acs.analchem.9b01325
复制
发表时间：
2019-07-02
影响因子：
7.4
通讯作者：
Weis, David D.
中科院分区：
文献类型：
--
作者：
Hageman, Tyler S.;Weis, David D.
差示氢交换质谱 (HX-MS) 测量对于识别蛋白质高级结构的差异非常有价值。通常，数据集很大，具有许多差异 HX 值，对应于在多个标记时间监测的许多肽。为了消除主观性并可靠地识别 HX-MS 测量中的显着差异，需要一种统计分析方法。在这项工作中，我们对麦芽糖结合蛋白和单克隆抗体英夫利昔单抗进行了无效 HX-MS 测量（即没有有意义的差异），以评估不同统计分析方法的可靠性。空测量对于直接评估与不同统计分析方法相关的风险（即错误地将差异分类为显着）和功效（即未能将真实差异分类为显着）非常有用。通过零测量，我们发现了常用方法的弱点。由于多重比较的问题，单独的显着性检验很容易出现假阳性。 Bonferroni 校正的结合导致检测范围大得令人无法接受，严重降低了功率。使用全局估计的显着性极限的分析方法也会导致检测极限的高估，从而导致功效损失。在这里，我们展示了一种基于火山图的混合统计分析，它将个体显着性测试与估计的全局显着性限制相结合，同时降低了误报的风险并保留了优越的功效。此外，我们强调了无效 HX-MS 测量的实用性，以明确评估用于将 FIX 差异分类为显着的标准。
Differential hydrogen exchange-mass spectrometry (HX-MS) measurements are valuable for identification of differences in the higher order structures of proteins. Typically, the data sets are large with many differential HX values corresponding to many peptides monitored at several labeling times. To eliminate subjectivity and reliably identify significant differences in HX-MS measurements, a statistical analysis approach is needed. In this work, we performed null HX-MS measurements (i.e., no meaningful differences) on maltose binding protein and infliximab, a monoclonal antibody, to evaluate the reliability of different statistical analysis approaches. Null measurements are useful for directly evaluating the risk (i.e., falsely classifying a difference as significant) and power (i.e., failing to classify a true difference as significant) associated with different statistical analysis approaches. With null measurements, we identified weaknesses in the approaches commonly used. Individual tests of significance were prone to false positives due to the problem of multiple comparisons. Incorporation of Bonferroni correction led to unacceptably large limits of detection, severely decreasing the power. Analysis methods using a globally estimated significance limit also led to an overestimation of the limit of detection, leading to a loss of power. Here, we demonstrate a hybrid statistical analysis, based on volcano plots, that combines individual significance testing with an estimated global significance limit, that simultaneously decreased the risk of false positives and retained superior power. Furthermore, we highlight the utility of null HX-MS measurements to explicitly evaluate the criteria used to classify a difference in FIX as significant.