MoDern-Cloud: An Artificial Intelligence Cloud for Accelerated NMR Spectroscopy

Multi-dimensional NMR spectroscopy is an invaluable biophysical tool. Non-uniform sampling is a powerful approach for shortening measurement time and increasing spectra resolution. Deep learning, a representative artificial intelligence technology, has shown astonishing potential in recovering spectra from undersampled NMR data. However, many existing problems, such as lack of robustness and explainability, greatly limit its applications. Here, we first devise the model-inspired deep learning framework (MoDern), which learns the optimal mapping from the undersampled data to the complete spectrum under a relatively comprehensible and low-computation-cost architecture. Second, we show that MoDern enables robust and high-fidelity spectra reconstruction for challenging multi-dimensional protein NMR and achieves high reliability on the relative concentration of the metabolite mixture. Third, we develop an easy-to-use cloud platform (MoDern-Cloud), to facilitate the widespread usage of this method, bridging the gap between high-performance and accessible implementations, and contributing a promising platform for further development of spectra analysis. These results suggest that MoDern-Cloud is a reliable, widely-available, understandable, and ultra-fast reconstruction technique for highly accelerated NMR.