Enabling Reliable Testing Of SMLM Datasets

荧光显微镜是生物医学研究的一项关键技术，因为它能够成像细胞中特定蛋白质的分布。在过去的二十年中，荧光显微镜的分辨率已经从200nm左右下降到20nm。这些方法中最流行的是单分子光显微镜（SMLM），它收集一系列图像，然后对其进行处理，以提高分辨率创建样品的重建图像。虽然这种方法在实验上很简单，并且可以获得非常好的分辨率，但它有一个主要缺点：它所依赖的图像处理可能导致重建图像中的伪影（虚假结构）难以发现，导致生物学家对他们的数据得出错误的结论。测试这些工件是极其困难的，并且直到最近才开发出这样做的方法。在这里，SMLM数据评估的两个主要开发人员想要联合起来创建一个新的工具，该工具可以使用他们的两种方法进行测试。我们将创建一个集成的软件包，用于测试分辨率（使用比所使用的标准精度更高的度量）和两种类型的工件。第一，人工锐化，是在多色和活细胞SMLM中遇到的常见问题，以及任何不找到荧光团位置的重建方法，而是修改输入图像。第二项研究考察了SMLM重建图像与样品的宽视场图像的一致性，并善于发现缺失的结构，在整个图像中检测到的荧光团比例的变化，以及背景引起的问题。通过对所有这些不同类型的工件进行测试，用户将能够确信他们的数据分析不会在图像中引入错误。代码将在Github存储库上开发，源代码将提供。这与申请人在发布开源软件方面的长期记录保持一致。该软件将以两种不同的软件包发布：ImageJ/Fiji插件（显微镜和生物/生物医学研究人员最常用的系统）和napari插件（一种新的基于python的图像处理软件包，越来越多的方法开发人员使用）。我们将寻求在发布之前将我们的方法建立为标准测试，并随着我们对SMLM的理解的提高，鼓励该领域开发进一步的扩展，并通过代码审查过程最终确定接受哪些修改。我们还将评估一系列公开可用的数据，并获得新的范例数据集，以允许我们向潜在用户说明可能的工件以及在什么情况下它们可能会遇到。虽然可用的SMLM数据越来越多，但这通常集中在其他人可能想要处理的高质量原始数据上。我们还希望生成数据集，其中有意生成可能产生错误的原始数据。缺乏一致的标准和测试导致了科学中的可重复性危机：不可能对SMLM研究中的数据分析进行基准测试或验证，因为没有这种研究应该遵守的商定标准。我们的方法结合了当前可用的最佳测试，得到了SMLM开发人员社区的支持和信任，也得到了依赖SMLM的设施管理人员和研究人员的支持。为测试设定这样的标准将允许期刊最终为SMLM数据验证提供一套一致的要求，这对于非专业研究人员来说实际上是可能实现的，并且提供了严格的质量检查。

Fluorescence microscopy is a key technology for biomedical research due to its ability to image the distribution of specific proteins in cells. Over the last twenty years the resolution achievable with fluorescence microscopy has been improved from around 200nm down to 20nm. The most popular of these methods, single molecule light microscopy (SMLM), collects a series of images and then processes it to create a reconstructed image of the sample with improved resolution. While this method is experimentally simple and can achieve very good resolution, it has a major drawback: the image processing it relies on can lead to artifacts (false structures) in the reconstructed image that are hard to spot, leading biologists to draw incorrect conclusions about their data. It is extremely difficult to test for these artifacts, and methods to do so have only recently been developed. Here the two leading developers of data assessment for SMLM want to combine forces to create a new tool which enables testing using both of their approaches.We will create an integrated software package that tests for resolution (using a metric improved in accuracy compared to the standard one used) and two types of artifact. The first, artificial sharpening, is common issue encountered in multicolour and live cell SMLM, and any reconstruction method that doesn't find positions of fluorophores, but instead modifies the input images. The second examines how well the SMLM reconstructed image agrees with a widefield image of the sample, and is good at spotting missing structure, variation in the proportion of fluorophores being detected across the image, and issues caused by the background. By testing for all of these different types of artifact, users will be able to have confidence that their data analysis is not introducing errors into their images.Code will be developed on a Github repository and source code will be made available. This is in keeping with the long track record of the applicants in releasing open source software. The software will be released in two different packages: as an ImageJ/Fiji plugin (the most common system used by microscopists and biological/biomedical researchers) and also in napari (a new python-based image processing package, which is increasingly used by methods developers). We will seek to establish our method as a standard test before publication, and encourage the field to develop further extensions as our understanding of SMLM improves, with a process of code review ultimately determining which modifications are accepted.We will also assess a range of publicly available data, and acquire new exemplar datasets to allow us to illustrate to potential users likely artifacts and under what circumstances they might be encountered. While there is an increasing amount of SMLM data available, this usually focuses on high quality raw data which others might want to process. We also want to generate datasets in which raw data likely to produce errors is deliberately generated. The lack of consistent standards and tests contributes to the reproducibility crisis in science: it is impossible to benchmark or validate the data analysis in SMLM studies because there are no agreed standards to which such studies should adhere. Our approach, uniting the best of the current tests available, has the support and confidence of the SMLM developer community and support from facility managers and researchers who rely on SMLM. Setting such a standard for testing will allow journals to finally put in place a consistent set of requirements for SMLM data verification, which is both actually possible for non-specialist researchers to fulfill, and provides rigorous quality checks.