Towards AI-enabled Fully Quantum (Bio)Molecular Simulations

Back

Alexandre Tkatchenko 

Department of Physics and Materials Science, the University of Luxembourg, Luxembourg 

The convergence between accurate quantum-mechanical (QM) models (and codes)  with efficient machine learning (ML) methods seem to promise a paradigm shift in  molecular simulations. Many challenging applications are now being tackled by  increasingly powerful QM/ML methodologies. These include modeling covalent  materials, molecules, molecular crystals, surfaces, and even whole proteins in explicit  water (https://www.science.org/doi/abs/10.1126/sciadv.adn4397). In this talk, I attempt  to provide a reality check on these recent advances and on the developments required  to enable fully quantum dynamics of complex functional (bio)molecular systems.  Multiple challenges are highlighted that should keep theorists in business for the  foreseeable future. 

Reference 1. Ensuring the accuracy of high-level QM methods (https://doi.org/10.1038/s41467-021-24119-3;  https://doi.org/10.1038/s41586-023-06587-3), 2. Describing intricate QM long-range interactions (https://doi.org/10.1126/sciadv.aax0024;  https://doi.org/10.1126/science.aae0509; https://doi.org/10.1103/PhysRevLett.128.106101), 3. Treating quantum electrodynamic effects that become relevant for complex molecules  (https://doi.org/10.1021/acs.jpclett.1c04222; https://doi.org/10.1103/PhysRevResearch.4.013011). 4. Developing increasingly accurate, efficient, scalable, and transferable ML architectures for  molecules and materials (https://doi.org/10.1038/s41467-022-31093-x; https://arxiv.org/abs/2209.14865;  https://arxiv.org/abs/2209.03985). 5. Accounting for the quantum nature of the nuclei and the influence of external environments  (https://doi.org/10.1038/s41467-020-20212-1; https://doi.org/10.1038/s41467-022-28461-y).I argue that  only a conjunction of all these developments will enable the long-held dream of fully quantum  (bio)molecular simulations.