Abstract
Variability in T cell performance presents a major challenge to adoptive cellular immunotherapy (ACT). This includes expansion of a small starting population into therapeutically effective numbers, which can fail due to differences between individuals and disease states. Intriguingly, modulating the mechanical stiffness of materials used to activate T cells can rescue subsequent expansion. However, the magnitude of this effect and the optimal stiffnesses differ between individuals, complicating the use of mechanosensing to improve cell production. The ability to predict this long-term, substrate-dependent expansion from a short-term assay would accelerate the deployment of immunotherapy. Here, it is demonstrated that short-term cell spreading predicts subsequent, mechanosensitive expansion. As an initial task, cell spreading is used to identify whether a sample of cells came from a healthy donor or a Chronic Lymphocytic Leukemia (CLL) patient. Notably, a deep learning (DL) model outperforms morphometric approaches to this classification task. This system also successfully predicts the long-term expansion potential of cells as a function of both source and mechanical stiffness of the activating substrate. By predicting long-term T cell function from small, diagnostic samples, this approach will improve the reliability and efficacy of cell production and immunotherapy.