Machine Learning for Materials Discovery
We develop machine learning methods to understand and design organic molecules and inoragnic materials. Our research focuses on learning structure–property relationships from large chemical and materials datasets using modern AI architectures such as graph neural networks, transformer-based models, and emerging foundation models for chemistry. These approaches enable the representation of complex molecular and materials structures and allow models to capture subtle chemical interactions that determine functional properties.
Beyond predictive modeling, we are particularly interested in generative models for inverse materials design. Techniques such as diffusion models, variational autoencoders, and autoregressive generative models enable the exploration of vast chemical design spaces and allow the creation of new molecules and materials with targeted properties. By combining generative AI with chemical knowledge and physical constraints, our goal is to develop intelligent design systems that can propose novel functional materials for applications in energy, catalysis, and advanced electronics.
Machine learning models that predict physical and chemical properties of molecules and materials directly from their structures, enabling rapid screening of large chemical spaces.
AI methods that evaluate whether proposed molecules and materials can realistically be synthesized, helping prioritize experimentally feasible candidates.
Generative AI models that design new molecules and materials with targeted properties by efficiently exploring vast chemical design spaces.
Selected papers
[1] Y. Kang#, H. Park#, B. Smit, and J. Kim*
A multi-modal pre-training transformer for universal transfer learning in metal–organic frameworks
Nature Machine Intelligence, 2023
[2] H. Park#, Y. Kang#, and J. Kim*
ACS Applied Materials & Interfaces, 2023
[3] S. Han#, Y. Kang#, T. Bae, V. Bernales, A. Aspuru-Guzik*, J. Kim*
Arxiv, 2026