Nadav Sellam

Nuclear Magnetic Resonance (NMR) spectroscopy provides rich, high-resolution insights into molecular structure and dynamics, but interpreting this data remains a complex and expert-driven process. This project aims to develop deep learning methods to bridge the gap between raw NMR data and molecular understanding by learning structure- and dynamics-aware representations from experimental spectra. Leveraging recent advances in geometric deep learning and equivariant models, we will explore architectures that can incorporate spatial and symmetry information intrinsic to molecular systems. A key objective is to integrate prior physical knowledge with data-driven approaches to improve robustness, generalization, and interpretability. The project will also investigate the use of generative models and uncertainty quantification to capture molecular heterogeneity and dynamics from NMR data. By aligning machine learning tools with the unique demands of NMR spectroscopy, this work seeks to reduce manual effort, increase throughput, and open new avenues for analyzing flexible or partially disordered biomolecular systems.