Schneider Magdalena C, Schütz Gerhard J
Institute of Applied Physics, TU Wien, Vienna, Austria.
Front Bioinform. 2022 Mar 4;2:811053. doi: 10.3389/fbinf.2022.811053. eCollection 2022.
The human mind shows extraordinary capability at recognizing patterns, while at the same time tending to underestimate the natural scope of random processes. Taken together, this easily misleads researchers in judging whether the observed characteristics of their data are of significance or just the outcome of random effects. One of the best tools to assess whether observed features fall into the scope of pure randomness is statistical significance testing, which quantifies the probability to falsely reject a chosen null hypothesis. The central parameter in this context is the -value, which can be calculated from the recorded data sets. In case of -values smaller than the level of significance, the null hypothesis is rejected, otherwise not. While significance testing has found widespread application in many sciences including the life sciences, it is hardly used in (bio-)physics. We propose here that significance testing provides an important and valid addendum to the toolbox of quantitative (single molecule) biology. It allows to support a quantitative judgement (the hypothesis) about the data set with a probabilistic assessment. In this manuscript we describe ways for obtaining valid -values in two selected applications of single molecule microscopy: (i) Nanoclustering in single molecule localization microscopy. Previously, we developed a method termed 2-CLASTA, which allows to calculate a valid -value for the null hypothesis of an underlying random distribution of molecules of interest while circumventing overcounting issues. Here, we present an extension to this approach, yielding a single overall -value for data pooled from multiple cells or experiments. (ii) Single molecule trajectories. Data from a single molecule trajectory are inherently correlated, thus prohibiting a direct analysis via conventional statistical tools. Here, we introduce a block permutation test, which yields a valid -value for the analysis and comparison of single molecule trajectory data. We exemplify the approach based on FRET trajectories.
人类思维在识别模式方面展现出非凡的能力，与此同时，往往会低估随机过程的自然范围。综合起来，这很容易在判断数据的观察特征是具有显著性还是仅仅是随机效应的结果时误导研究人员。评估观察到的特征是否属于纯随机范围的最佳工具之一是统计显著性检验，它量化了错误拒绝所选零假设的概率。在此背景下的核心参数是p值，它可以从记录的数据集中计算得出。如果p值小于显著性水平，则拒绝零假设，否则不拒绝。虽然显著性检验在包括生命科学在内的许多科学领域都有广泛应用，但在（生物）物理学中却很少使用。我们在此提出，显著性检验为定量（单分子）生物学的工具箱提供了一个重要且有效的补充。它允许通过概率评估来支持对数据集的定量判断（假设）。在本手稿中，我们描述了在单分子显微镜的两个选定应用中获得有效p值的方法：（i）单分子定位显微镜中的纳米簇聚。此前，我们开发了一种称为2-CLASTA的方法，该方法能够在规避重复计数问题的同时，为感兴趣分子的潜在随机分布的零假设计算有效p值。在此，我们展示了该方法的扩展，为从多个细胞或实验汇总的数据得出单个总体p值。（ii）单分子轨迹。单分子轨迹的数据本质上是相关的，因此禁止通过传统统计工具进行直接分析。在此，我们引入一种块排列检验，它为单分子轨迹数据的分析和比较产生有效p值。我们以FRET轨迹为例说明了该方法。
