(24a) Invited Talk: Machine Learning Assisted Design of Immunomodulatory Biomaterials

Artificial antigen-presenting cells (aAPCs) have emerged as promising tools for ex vivo activation and expansion of T cells, offering potential for scalable and safe manufacturing of cellular immunotherapies. However, current design of aAPC platforms fall short of their full potential due to their limited ability to mimic the natural microenvironment provided by APCs in vivo. This limitation is primarily attributed to lack of comprehensive understanding of interplay of signalling cues involved in immune cells activation.

Here, biodegradable hydrogel microbeads coated with a supporting lipid bilayer is created as a versatile APC-mimetic platform for systematic investigation of important cell-mimetic features and for combination screening of T cell stimulating molecules. The modular design of our platform allows for convenient manipulation of various physical features and facilitates the incorporation of novel combinations of stimulatory and co-stimulatory cues in a plug-and-play manner, allowing for high-throughput screening of vast array of signalling molecules combinations. To aid in this investigation, we employed a combination of advanced machine learning techniques, including neural network and self-validating ensemble model to explore the design principles that dictate the potency of aAPC-microbeads and predict the optimal bead design tailored to specific cell types of interest.

To generate training data for our model, we synthesized a library of beads exploring large design space featuring 8 design parameters, including microbead stiffness, size, number of beads, surface signal combinations, densities, and molar ratios. These beads were co-cultured with peripheral blood mononuclear cells and monitored for 5 and 10 days. Cell expansion, phenotypic profile, and cytotoxicity are selected as screening outcome measurements, utilizing fluorometric cell counting assays, multiwell-plate-based flow cytometry, and tracking of fluorescently-labeled target cell survival.

Through our systematic exploration, we have gained valuable insights into optimizing the design of aAPC-microbeads to enhance T cell functionality and therapeutic efficacy. Our findings contribute to the advancement of cellular immunotherapies by providing a better understanding of the factors that influence T cell stimulation and paving the way for the development of improved aAPC platforms.