Quantitative Phase Imaging is currently used to count and determine viability of a variety of cell types. When a cell culture process involves an infection step (e.g. Baculovirus Expression Vector System in insect cells), it is not yet possible to continuously monitor the infection kinetics. To do so, sampling and off-line analysis are required. However, real time results are important because they can impact the way the culture is performed, for instance addition of nutrients, change of cell culture parameters, optimal harvest time, etc.
Our study shows that using a novel technology, Quantitative Phase Imaging (based on Differential Digital Holographic Microscopy), a detection system can be trained to identify infected cells, and then be used to determine the percentage of infected cells within the culture in real-time. First, the system was trained using two limited data sets, one population made of non-infected cells only (typically at the beginning of the cell culture process) and one population made of infected cells only. Based on a random forest tree (machine learning that creates algorithms for classification), it has been possible to design a detection algorithm that is capable of identifying cells that are infected. With these limited data sets, a false positive rate of 8.2% and a false negative rate of 7.3% were obtained. Following this, the algorithm was integrated into OsOne, a cell culture monitoring software tool, and used on different Sf9 cultures infected with the Baculovirus Expression Vector System. Results have shown that a strong correlation between offline analysis and online analysis can be achieved, confirming that the detection algorithm created is reliable and in line with standard, off-line methods.
In conclusion, the detection system was trained to extract a specific cellular fingerprint pre- and post-infection, thus allowing in-line, continuous monitoring of an infection process.