Three Pillars of AI-Driven Discovery

Our Bioinformatics group sits at the frontier of AI-driven molecular science. We develop and apply machine learning methods — with a strong emphasis on diffusion models and generative AI — to accelerate molecular discovery, de novo drug design, and predictive modeling of biological systems. Research directions include ligand-based virtual screening, structure-based molecular docking, scaffold-based generative design, and binding affinity prediction. A central thread across all projects is explainability: we build models that not only predict but also illuminate the structural and chemical reasoning behind their outputs, supporting downstream validation and trust by domain scientists.

Time series data is ubiquitous across science and industry, yet remains challenging due to noise, irregular sampling, and complex temporal dependencies. Our group addresses these challenges across the full research spectrum — from building general-purpose foundation models for time series to exploring practical downstream applications such as data imputation via retrieval-augmented generation (RAG), forecasting through visual representation, and online learning in streaming environments. We actively investigate multimodal time series analysis that fuses numerical signals with contextual information, as well as spatio-temporal modeling for geospatial and sensor-network data. Future research directions include text-based approaches to time series understanding, where language models are adapted to reason over sequential numerical patterns.

Material design is a combinatorially vast challenge that AI is uniquely positioned to accelerate. Our Materials group applies deep learning, graph-based property prediction, and generative models to guide the design of novel materials with precisely tuned functional properties. Current application focuses include: (1) thermally adaptive materials for reducing heat accumulation in greenhouse structures, directly supporting sustainable agriculture; (2) spectrally selective materials for camouflage and stealth applications in military vehicle systems, requiring precise control of electromagnetic response across optical and infrared ranges; and (3) biocompatible functional materials for biosensor platforms, where sensitivity, selectivity, and biocompatibility must be jointly optimized. Across all three directions, our approach combines physics-informed modeling with data-driven generative methods to navigate the materials design space efficiently.