High-throughput virus quantification using cytopathic effect area analysis by deep learning
Traditional infectivity-based virion quantification methods, such as 50% tissue culture infectious dose (TCID50) and plaque assays, are typically performed in 6-, 12-, 24-, or 48-well cell culture plates and require manual work and analysis based on legacy protocols. Adaptation of these methods to high-throughput formats (96-, 384- or 1,536-well plates) is challenging due to both assay automation constraints and the lack of well surface area available for reliable analysis. Here, we present a scalable alternative to traditional methods that uses whole-well image thresholding to quantify infectious virions by measuring virus-induced cytopathic effect (CPE) via cell lysis and detachment. The CPE area assay is best positioned as a practical high-throughput preliminary screening tool, effectively quantifying samples within the assay range and flagging samples above or below the concentration thresholds. To improve analysis efficiency and reduce user bias in CPE area selection, we evaluated an nnU-Net model for automated image segmentation against images segmented by manually defined brightness thresholds. The model achieved perfect correlation with manual thresholding (R2=1.00), showing minimal differences in the identified CPE area and thus validating nnU-Net as a reliable alternative to manual analysis. This approach provides two complementary pipelines: manual thresholding, which tolerates adjustments of hyperparameters, and fully automated segmentation via nnU-Net, which streamlines analysis and enhances throughput. This flexible CPE area assay enables accurate and automated quantification in high-throughput screening formats, thereby greatly accelerating a routine laboratory task while decreasing subjectivity and bias. Author summaryMeasuring the amount of virus in a biosample is a critical part of viral research and testing of new treatments. Traditional methods are slow, labor-intensive, and not easily scaled up to handle large numbers of samples. In this study, we developed a new approach that makes the process faster, more efficient, scalable, and less reliant on manual steps. We used microscopy imaging to capture how cells respond to virus infections and analyzed these images automatically to identify and measure areas where the virus had damaged cells. The analysis pipeline allows researchers to fine-tune settings or run the process using machine learning. This method is flexible and adaptable to a variety of viruses and experiments. By increasing the number of samples that can be tested at once while reducing the time and effort needed for analysis, our approach has the potential to accelerate research in virology, drug development, and public health.