LAUSR

Ghost cytometry (GC) is a recently realized approach for machine learning-driven analysis of image information without image production and enabled high throughput cell sorting through its integration with a microfluidic sorting technology. When a supervised machine learning is adopted in GC, a classifier model is first trained using a data set comprising of simultaneously measured imaging waveform signals and cell specific (ground truth) labels indicating cell types and functions. This trained model, in turn, is able to infer the cell specific label directly from the imaging waveforms at a speed of several orders of magnitude faster than the other methods, thus enabling subsequent cell sorting at high speed. Here, we investigate the applicability of label-free GC to manufacture and isolate the effective therapeutic cells.First, we tested classification of live T cells from dead cells to understand the potential of GC for monitoring the healthiness of live cells in cell manufacturing processes. A model for classifying the label-free information was trained using labels based on the positivity of annexin V and PI staining and applied to test data. Next, we assessed classification of T cells from live blood samples after hemolysis process. In this experiment, labels of a CD3 marker was used to construct a training data set to develop the T cell classifier. Finally, we applied the label-free GC to identify cells with low glycolytic activity because the cell metabolic status has a big impact on their therapeutic potency and persistency. Selection of the low glycolytic T cell may address current challenges on CAR-T cell therapy such as sustained tumor elimination. As a result, GC was found to classify not only live and dead cells but also apoptotic cells. Moreover, GC was able to discriminate CD3 positive T cells from PBMC with a good accuracy. These results indicate that our label-free GC can detect the healthy live cells without using any surface markers and could be beneficial for monitoring the quality of T cell products. Furthermore, the label-free GC may have a potential to classify the cells based on their phenotypic properties such as metabolic status. We would like to share our recent data at the meeting and discuss the potential of label-free GC for delivering sophisticated therapeutic cells, especially CAR-T cells, with good efficacy and persistency. Citation Format: Hiroshi Ochiai, Keiki Sugimoto. Machine learning-driven label-free cell sorting for cell therapy [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 4234.