A BUFFER REGION QM/MM APPROACH - BURNN FOR BIOMOLECULAR SIMULATIONS OF PROTEIN-LIGAND COMPLEXES AND METALLOORGANIC MOLECULES

Computational chemistry has made significant progress in studying biomolecular properties over the last few decades. Molecular simulations provide a detailed examination of various macroscopic thermodynamic and structural properties. However, the accuracy of these calculated properties heavily depends on the quality of the empirical force field (MM) used to evaluate the underlying interactions. Empirical force fields, with their simplistic approach of assigning partial point charges to atoms, often fail to describe many nonclassical phenomena such as charge transfer, polarization or excited states. As a result, a quantum mechanical (QM) description becomes inevitable. Unfortunately, the computational effort required for QM calculations is immense, making continuous trajectories at the QM level unfeasible even for small systems. To address this challenge, hybrid QM/MM techniques have been developed. These methods bridge the gap between accuracy and time efficiency by enabling a precise QM-level description for a critical small part of a system, known as the inner region. More recently, machine-learned interatomic potentials (MLIPs) have demonstrated the ability to learn the potential energy surface at the underlying reference QM level, providing a significant speed-up in calculations. The BuRNN methodology seeks to further enhance traditional QM/MM techniques by transitioning to the MLIP/MM level, resulting in speed-ups by orders of magnitude, and by introducing an additional buffer region that is calculated at both levels of theory to reduce artifacts at the QM/MM interface. Since BuRNN has been shown to work for a hexaaquairon(III) complex in water, our first aim is to further develop and evaluate the BuRNN methodology in increasingly complex systems, such as protein-ligand systems and metalloorganic compounds. Our second aim involves implementing a robust workflow to generate MLIPs for complex systems by incorporating enhanced sampling techniques at the MLIP level. Finally, we aim to advance relative binding free-energy calculations by allowing perturbations at the MLIP level, thereby improving protein-ligand binding affinity predictions and enabling accurate calculations of free energies for coordinative binding in metalloorganic complexes.