De Trina, Thangamani Subasini, Urbański Adrian, Yakimovich Artur
Center for Advanced Systems Understanding (CASUS), Görlitz, 02826, Germany.
Helmholtz-Zentrum Dresden-Rossendorf e. V. (HZDR), Dresden, 01328, Germany.
Sci Data. 2025 Apr 30;12(1):719. doi: 10.1038/s41597-025-05030-8.
Virological plaque assay is the major method of detecting and quantifying infectious viruses in research and diagnostic samples. Furthermore, viral plaque phenotypes contain information about the life cycle and spreading mechanism of the virus forming them. While some modernisations have been proposed, the conventional assay typically involves manual quantification of plaque phenotypes, which is both laborious and time-consuming. Here, we present an annotated dataset of digital photographs of plaque assay plates of Vaccinia virus - a prototypic propoxvirus. We demonstrate how analysis of these plates can be performed using deep learning by training models based on the leading architecture for biomedical instance segmentation - StarDist. Finally, we show that the entire analysis can be achieved in a single step by HydraStarDist - the modified architecture we propose.
病毒空斑测定法是在研究和诊断样本中检测和定量感染性病毒的主要方法。此外，病毒空斑表型包含有关形成它们的病毒的生命周期和传播机制的信息。虽然已经提出了一些改进方法，但传统测定法通常涉及对空斑表型的手动定量，这既费力又耗时。在这里，我们展示了痘苗病毒（一种典型的痘病毒）空斑测定板数字照片的注释数据集。我们通过基于生物医学实例分割的领先架构——StarDist训练模型，演示了如何使用深度学习对这些平板进行分析。最后，我们表明通过HydraStarDist（我们提出的改进架构）可以在单个步骤中完成整个分析。
