A team of Johns Hopkins engineers has leveraged a machine learning model to identify lipid nanoparticle (LNP) formulations that efficiently deliver gene-editing proteins into cells, marking a key advance toward safer and more effective gene and cell therapies.

Led by Hai-Quan Mao, director and core researcher at the Johns Hopkins Institute for NanoBioTechnology and professor of materials science and engineering at the Whiting School of Engineering, this team conducted high-throughput screening (HTS) of hundreds of LNP formulations to evaluate their ability to deliver gene-editing proteins into cells. The team then applied machine learning analysis to extract key formulation features that enhance in vitro gene editing efficiency.

“By combining a high-throughput platform with machine learning, we can predict which LNP formulations are most effective for protein delivery,” said Xiaoya Lu and Yining Zhu, PhD candidates in Materials Science and Engineering and Biomedical Engineering and the study’s lead authors. “This data-driven approach accelerates the discovery and optimization of nanoparticle formulations for gene-editing applications.”

With the success of LNPs in mRNA vaccines, these carriers have become a cornerstone of nucleic acid delivery. Unlike mRNA delivery, which relies on cellular translation of genetic instructions, protein delivery offers an alternative strategy that enables more direct and rapid genome editing without requiring transcription or translation inside target cells.  Optimizing LNPs for protein delivery still remains a major challenge due to limited understanding of how formulation composition affects transfection efficiency.

One particularly valuable but difficult application of gene-editing therapeutics is in vivo T cell engineering. T cells play a critical role in adaptive immunity and are critical targets for treating infectious diseases, autoimmunity, and cancer. Traditional approaches involve isolating T cells from a patient, genetically modifying them, expanding them in culture, and reinfusing them back into the body. This process is labor intensive, costly, and constrained by patient variables that limit its scalability and accessibility.

Direct in vivo gene editing of T cells also offers advantages, including greater scalability and the potential for broader clinical applications. However, achieving efficient in vivo gene editing in T cells remains a major challenge due to their dynamic circulation, complex microenvironment, and the need for selective targeting. Overcoming these challenges would enable more accessible and effective immunotherapies.

Using one of the optimized LNP formulations identified through this approach, the researchers achieved targeted knockout of specific genes, CCR5 and PD-1, in splenic T cells, demonstrating potential applications in HIV resistance and cancer immunotherapy.

“This work establishes a framework for machine learning–guided optimization of LNPs for protein delivery,” said Mao. “It represents an important step toward developing efficient, non-viral platforms for in vivo gene editing.”

The team’s findings were published in Science Advances.

By integrating data science with materials engineering, the study provides new insights into how rational design and predictive modeling can potentially accelerate the development of next-generation delivery systems for precise and safe genome engineering.