[Related research] "Physics-Informed Neural Network (PINN)-based multiscale and multiphysics design system for Sodium-ion batteries", Mid-career Research Grant from NRF of Korea 2025-2029.

Sodium-ion batteries (SIBs) are gaining attention as a sustainable alternative to lithium-based systems, thanks to sodium’s high abundance and low cost. However, achieving high performance and stability remains a key challenge.

Our group develops a Physics-Informed Neural Network (PINN)-based multiscale design framework to optimize SIBs. We integrate DFT-based material data, neural networks, and multiphysics simulations to capture stress, heat, and ion transport across scales. This enables efficient battery design under real-world constraints, such as temperature and mechanical stress.

Our group applies artificial intelligence to design and optimize biochip systems for biomedical applications, for example exosome separation and 3D spheroid culture. 

In the exosome separation platform, we combine multiphysics simulation and Bayesian optimization to refine microfluidic chamber and outlet geometries. This approach enables precise control over flow conditions and particle behavior, significantly improving exosome separation efficiency and yield.

In parallel, we apply deep learning models to the design of 3D spheroid culture chips for organ-on-a-chip applications. Through multiphysics simulation-based data generation and neural network training, we develop predictive tools that relate device geometry to key biological outcomes—such as spheroid formation, nutrient flow, and drug response. This approach allows rapid evaluation and fine-tuning of chip designs for high-performance drug screening platforms.

By integrating simulation and machine learning, we provide a data-driven framework for rapidly developing customized biochips suited for diagnostics, drug screening, and personalized therapy research.

Next-generation mRNA vaccines require both efficient delivery systems and rational antigen design to achieve high efficacy and stability. Our group develops AI-assisted frameworks to address both challenges through data-driven optimization and molecular modeling.

To guide the formulation of lipid nanoparticles (LNPs) for mRNA delivery, we integrate microfluidic mixing data with predictive machine learning models. This enables rapid exploration of formulation parameters and mixing conditions to enhance encapsulation efficiency and particle uniformity.

In parallel, we apply generative AI and molecular simulation techniques to design and screen novel antigen sequences. By coupling large-scale sequence generation with structural stability and binding energy analysis, we identify optimized candidates with improved immunogenic potential.

Together, these approaches support an intelligent, simulation-guided pipeline for the development of mRNA-based vaccine systems.