Single-cell viability assessment using YOLOv8n object detection

Assessing cellular viability is a fundamental step for high-throughput screening (HTS) in drug development, particularly for evaluating the unintended adverse toxicity of standard medications or, conversely, the therapeutic efficacy of oncological agents, in which inducing cell death is the desired outcome. Conventional methods typically rely on ATP-dependent enzymatic substrate conversion assays to measure cellular metabolic activity as a proxy for viability. While effective, these plate reader-based methods provide only bulk measurements, limiting the level of detail in the analysis. Recent advancements in convolutional neural networks for biological image analysis have introduced promising alternatives, allowing the assessment of cellular viability at single-cell resolution. This approach provides more precise and detailed viability assessments than bulk assays. In this study, we present a methodology employing a trained YOLOv8n detection model, developed by Ultralytics, for endpoint and kinetic assays. This method offers rapid and precise toxicity predictions, making it a valuable tool in drug discovery.