Machine Learning-Driven Discovery of a Lipid Nanoparticle for In-Vivo T-Cell Transfection in Non-Human Primates

The limited availability of in-vivo transfection of T-cells with mRNA therapeutics remains a major bottleneck in the development of scalable and accessible gene and cell therapies​​. Lipid nanoparticles (LNPs) offer an in vivo, non-viral alternative to ex-vivo genetic engineering but have historically shown poor performance in T-cells​. Our machine learning approach enabled the rapid design of novel LNPs, seamlessly integrating in-silico prediction with wet-lab validation to accelerate the discovery and optimization process​.Here, we report the machine learning (ML)-guided discovery of FMB-3199, a passively targeted LNP capable of safe in vivo T-cell delivery without surface-conjugated ligands or antibodies, identified through iterative design-build-test-learn (DBTL) cycles that progressively refined and improved the model’s predictive quality. In NSG mice injected with human peripheral blood mononuclear cells (hPBMCs), FMB-3199 achieved ~ 60% transfection of human T-cells in vivo, further validating its translational potential. In addition, its analogs achieved up to 98% killing of NALM6 cells within 48 hours in vitro, underscoring their functional therapeutic efficacy. Finally, in non-human primates (NHPs), FMB-3199 enabled dose-dependent safe CD3⁺ T-cell transfection (~ 2.5–15%), with ~ 25% in CD4⁺ T-cells while minimizing liver uptake.Together, these findings establish a scalable and generalizable platform for in-vivo T-cell engineering, accelerating the development of next-generation mRNA-based cell therapies.