(1330-C) Label-free morphological profiling and isolation of immune cell subsets using VisionSort, a novel, AI-based flow cytometry platform

Abstract: Identification, characterization, and minimally invasive isolation of specific populations of human immune cells are critical for understanding and treating disease. Modern cellular immunotherapy approaches demand innovative technologies that can isolate specific immune cell subsets in their native biological/functional states to improve the quality and efficacy of cell therapy drug products. In addition, novel approaches that can assess subtle phenotypic differences in immune cells are needed to advance basic life science research and development (R&D).

Here we used VisionSort, a new label-free, artificial intelligence (AI)-driven cellular analysis and sorting platform, to isolate and characterize truly untouched human immune cell subsets for downstream R&D applications. We present data on label-free identification of three immune cell subsets using morphological profiling. In the first, we used VisionSort to distinguish between T cell activation states label-free, an emerging need in cell therapy R&D. By capturing single-cell digital phenotypes, we characterized mouse T-cells and generated ‘ground truth’ functional profiles for activated and non-activated T cells. A set of machine-learning derived classifiers was generated to identify these phenotypic classes in unlabeled T-cell subsets. The classifier showed an area under the curve (AUC) performance for differentiating between phenotypically defined T cell populations of 0.917. In addition, by using unsupervised machine learning, we were able to resolve activated and non-activated T cell populations label free, using morphological data alone.

Next, we used VisionSort for label-free isolation of plasma B-cells, a critical need in therapeutic antibody discovery and development programs. Human B cells were cultured under conditions that promoted either B cell activation or plasma cell differentiation. IgD, CD38, and CD27 were used as ‘ground truth’ markers to define populations of B cells and plasma cells. The VisionSort-derived classifier built using supervised machine learning showed excellent classification of B cells and plasma cells with an AUC score of 0.941.

Last, we used VisionSort to characterize macrophage subsets. From human peripheral blood monocytes, we induced M1 and M2 polarized macrophages in vitro and generated machine learning classifiers to recognize these populations on VisionSort using supervised machine learning. The resulting classifier had high discriminatory power and reproducibility when applied to test samples with an AUC of 0.878 +/- 0.002 (n=6).

In conclusion, here we report results on the use of a novel, label-free cytometry platform to characterize and isolate human immune cell subsets using morphological profiling and AI. The approach enables label-free isolation of target immune cell subsets with defined phenotypic profiles in an unperturbed state and has practical applications for investigators in basic life sciences as well as drug developers in small molecule, antibody, and cell therapy R&D.