(1137-B) Label-free High-Throughput Viability Analysis using Advanced Machine Learning Algorithms
Monday, February 5, 2024
2:00 PM – 3:00 PM EST
Location: Exhibit Halls AB
Abstract: Cell viability assays are essential to determine the toxicity of compounds or cellular health during the cell culture routine in various research fields like drug discovery, cancer research, bioprocessing, or cell line development. One of the most common methods to access cell viability is the well-established Trypan Blue staining. The cell suspension is mixed with the Trypan Blue dye and is visually examined under a light microscope to determine whether the cells include or exclude the dye. However, Trypan Blue is toxic to the cells at higher concentrations or prolonged exposure. Consequently, it is an endpoint assay, and false results can be obtained if the analysis takes longer than expected. Moreover, the staining step adds unwanted complexity to automated liquid handling processes. Despite that, Trypan Blue is classified as a carcinogen, which makes proper handling and disposal of the toxic dye difficult. Thus, there is a need to explore alternative methods to overcome these obstacles, while being reliable, time- and cost-effective. Therefore, we aimed to develop a solution using brightfield images of unlabeled cells to assess the cell viability of suspension cells using algorithms of machine learning/artificial intelligence (AI). To do so, we generated training data by staining a mixed population of live and dead CHO-K1 or HEK293T cells with Hoechst 33342, Calcein-AM, and Propidium Iodide (PI) and imaged the cells with our high-content imager CELLAVISTA 4K. Based on this staining, the cells were labeled and assigned to the class live (Calcein-AM positive/PI negative) or dead (PI positive) using our in-house software AI-Studio+. The AI models learned from the input data to detect cells and assess their viability based on morphological differences. Subsequently, we validated the accuracy of the AI model by comparing its results with those obtained from the classical image processing of a Trypan Blue staining. Both assays showed similar results regarding viability and cell count. Moreover, the standard deviation was lower for the AI-analyzed data than for the classical Trypan Blue assay. The AI model can be integrated into our YT-SOFTWARE so that imaging and analysis can be performed within the same software ecosystem reducing time and costs for additional training and support. Taken together we developed a reliable tool to evaluate the cell viability of suspension cells in an automated high-throughput manner, without the necessity for toxic dyes and incubation steps. On the one hand, this is interesting for low-volume fermentation in microplates and, on the other hand, the samples tested for viability and cell count can be used directly 1:1 for subsequent analysis.