Accurate and efficient molecular spectra simulation using ML force field and GNN
This solution leverages machine learning and graph neural networks to simulate molecular spectra efficiently. It integrates ML force fields with quantum chemistry, reducing computational costs while maintaining accuracy.
Overview
This research initiative focuses on developing a cutting-edge solution for simulating molecular spectra with enhanced accuracy and efficiency. By utilizing machine learning force fields (ML-FF) alongside graph neural networks (GNN), the project aims to streamline the prediction of molecular spectral properties such as IR, Raman, UV-Vis, and NMR spectra. This approach promises to significantly reduce computational costs compared to traditional quantum chemistry methods while maintaining high precision.
Technical specifications
- Deep Equivariant Tensor Attention Network (DetaNet): A novel network designed to predict molecular tensorial properties accurately. Initially trained on a limited dataset, future plans include expanding the dataset to enhance generalization.
- Machine Learning Force Field (ML-FF): Integrates ring polymer molecular dynamics and autocorrelation functions to simulate molecular dynamics spectra efficiently, overcoming traditional computational limitations.
- Graph Neural Network (GNN): Capable of translating SMILES representations directly into molecular spectra, bypassing complex geometrical evolutions. This requires a comprehensive and precise experimental spectra database for optimal accuracy.
Technology readiness level
This technology is currently at TRL 3, indicating that it has been validated in laboratory settings but requires further development and dataset expansion to enhance its applicability across a broader range of chemical elements and functional groups.
About USTC
USTC is a comprehensive public research university in Hefei with additional research hubs in the Yangtze River Delta, distinguished by intensive graduate training and close affiliation with the Chinese Academy of Sciences. Industry partners engage through co‑located national laboratories and large‑scale user facilities, a nearby research and technology park, and joint university–CAS platforms that streamline collaboration. Research is funded by competitive national programs, including the National Natural Science Foundation of China and the Ministry of Science and Technology, with added provincial and municipal support. A dedicated technology transfer office supports IP, licensing, and startup formation.
delivered to your inbox.