128 Label-free morphometric characterization of T cells for cell and gene therapy research and development

<h3>Background</h3> Identification, characterization, and minimally invasive isolation of T cells is critical for developing cellular therapies. <h3>Methods</h3> Here we used VisionSort, an artificial intelligence (AI)-driven cellular analysis and sorting platform, to characterize and isolate human T cell subsets with therapeutically relevant phenotypes, label-free. By capturing single-cell digital phenotypes, we characterized human T-cells and generated ‘ground truth’ functional profiles for 1) glycolysis level, 2) exhaustion state 3) activation or resting profiles and 4) viability. A set of machine-learning derived classifiers was generated to identify these phenotypic classes label-free. <h3>Results</h3> The classifiers showed area under the curve (AUC) performance ranges for detecting specific, phenotypically defined T-cell populations between 0.923 and 0.995. Using a combination of supervised and unsupervised machine learning approaches, we also show identification of follicular helper T (Tfh) cells from other CD4+ T cells based on morphology alone. When we analyzed the morphological profiles of T cell subsets by unsupervised machine learning using uniform manifold projection and approximation (UMAP), we found that both gamma delta (γδ) T cells were morphologically distinct from alpha beta (αβ) T cells and CD8+ T cells were distinct from CD4+ T cells. <h3>Conclusions</h3> Here we report on the use of AI-based cytometry to characterize and isolate phenotypically defined T cell subsets label-free, with applications in basic life science and cell therapy R&amp;D (figure 1).