FTMS Laboratory for Human Health Research, Department of Chemistry, North Carolina State University, Raleigh, North Carolina 27695, United States.
J Proteome Res. 2024 Oct 4;23(10):4384-4391. doi: 10.1021/acs.jproteome.4c00360. Epub 2024 Sep 3.
Quality control and system suitability testing are vital protocols implemented to ensure the repeatability and reproducibility of data in mass spectrometry investigations. However, mass spectrometry imaging (MSI) analyses present added complexity since both chemical and spatial information are measured. Herein, we employ various machine learning algorithms and a novel quality control mixture to classify the working conditions of an MSI platform. Each algorithm was evaluated in terms of its performance on unseen data, validated with negative control data sets to rule out confounding variables or chance agreement, and utilized to determine the necessary sample size to achieve a high level of accurate classifications. In this work, a robust machine learning workflow was established where models could accurately classify the instrument condition as clean or compromised based on data metrics extracted from the analyzed quality control sample. This work highlights the power of machine learning to recognize complex patterns in MSI data and use those relationships to perform a system suitability test for MSI platforms.
质量控制和系统适用性测试是至关重要的协议，旨在确保质谱研究中数据的可重复性和再现性。然而，由于同时测量化学和空间信息，质谱成像（MSI）分析带来了额外的复杂性。在此，我们采用了各种机器学习算法和一种新型的质量控制混合物来对 MSI 平台的工作条件进行分类。根据每个算法在未见数据上的性能进行评估，使用阴性对照数据集进行验证，以排除混杂变量或偶然一致的情况，并利用该算法来确定实现高精度分类所需的样本量。在这项工作中，建立了一个强大的机器学习工作流程，该流程可以根据从分析的质量控制样本中提取的数据指标，准确地将仪器状态分类为干净或受损。这项工作强调了机器学习在识别 MSI 数据中的复杂模式并利用这些关系来对 MSI 平台进行系统适用性测试方面的强大功能。
