Research
1. AI-Assisted Chromatographic Separation & Purification, and Automation Platforms
Chromatographic separation is an indispensable core tool in chemical research, yet its method development has long depended on personal experience and trial-and-error — inefficient and poorly reproducible. We are committed to deeply integrating AI with laboratory automation, transforming chromatography from an intuition-dependent “craft” into a predictable, programmable “science”.
By acquiring standardized, large-scale chromatography data through automated robotic platforms and developing machine-learning models with embedded mechanistic constraints (such as QGeoGNN), we achieve cross-scale, interpretable prediction from thin-layer chromatography (TLC) to high-performance liquid chromatography (HPLC). Combined with transfer learning and condition embeddings, the models generalize across different labs, instruments, and columns, providing efficient, reliable, and reproducible separation solutions for chemical research.
2. AI Molecular Spectroscopy
Molecular spectroscopy (infrared, UV, mass, and circular dichroism) is a key approach to decoding molecular structure and properties, yet acquiring such data has long relied on time-consuming, expensive theoretical calculations or experiments. We build large-scale, high-quality spectral datasets and design architectures such as graph neural networks and Transformers to learn the complex mapping between molecular structure and spectra, achieving second-level, accurate prediction of multiple spectra from molecular topology. Through strategies such as decoupling peak attributes to improve interpretability, we provide high-throughput analysis tools for chiral recognition, drug screening, and materials design.
3. Physical AI for Chemistry
Traditional chemical discovery relies on people working hands-on in the lab — planning routes, assembling setups, observing reactions, and iterating manually, which is slow and hard to scale. We extend artificial intelligence into the physical world through Physical AI: AI agents operate robots to autonomously carry out synthesis and discovery experiments in real laboratories, closing a perceive–reason–act–feedback loop. The agents perceive reaction progress in real time and decide the next step, while robots precisely execute operations such as dosing, temperature control, separation, and characterization — moving chemical discovery from “people doing in the lab” to “AI doing in the lab”, building a closed loop for autonomous chemical discovery.