Talk and poster at ESPA2026 in Salamanca

Together with our colleagues from TunonLab, we made a road trip to Salamanca for the 14th Electronic Structure, Principles and Applications conference (ESPA2026), held from 30 June to 3 July together with the biennial meeting of the Chemistry and Computation group of the Spanish Royal Society of Chemistry.

Kirill gave the talk Unexpected advantage of physical models in hybrid Machine Learning / Molecular Mechanics simulations. He showed that the EMLE scheme, trained only on gas-phase data, goes beyond mimicking QM/MM: it predicts Raman spectra, avoids the charge spill-out problem of QM/MM, and enables enhanced sampling methods that are impossible with standard QM/MM or naïve ML/MM potentials.

Rubén presented the poster Accelerating convergence of ML/MM simulations with mean field approach. Because EMLE separates the gas-phase energy of the ML region from its interaction with the environment, the expensive ML energy can be computed once while the MM environment relaxes around it. The averaged MM forces then act as a mean field on the reactive region, giving free energy profiles at a fraction of the cost of conventional ML/MM simulations.