Abstract
Cancer is among the leading causes of death worldwide, underscoring the need for novel therapeutic strategies. Hematologic cancers are among the most common and aggressive malignancies, often require intensive chemotherapies or bone marrow transplants. These treatments have significant side effects and often fail in advanced stages. CAR-T cell therapy is a revolutionary new type of immunotherapy where T cells are engineered to express Chimeric Antigen Receptors. Despite impressive clinical outcomes, challenges persist, including suboptimal design stemming from trial-and-error approaches and domains being studied in isolation. This project introduces a data-driven approach that integrates generative deep learning, predictive modeling, and experimental validation in a closed-loop framework for more efficient CAR design. We first assemble a CAR-T database to fine-tune a generative model, trained on general protein sequences, allowing for the generation of de novo CAR constructs. A predictive model evaluates these constructs and guides the selection of lead candidates for in vitro testing. Experimental feedback is continuously fed back into the generative model, refining its ability to produce safer, more effective CARs. By combining expertise from the Laboratory of Experimental Hematology and the Adrem Data Lab, this interdisciplinary project aims to deepen our understanding of CAR domain interactions and accelerate the development of next-generation CAR-T cell therapies.
Researcher(s)
Research team(s)
Project type(s)