Learning chemistry with multiple first-principles simulations

Abstract

Huge parallel high-performance computing (HPC) architectures are today available laboratories for modelling atomic forces with high accuracy and for large samples of atoms. Modern statistical tools allow to simulate the statistics of these samples, while first-principles molecular dynamics (MD) can probe the interactions within large atomic samples, including possible chemical reactions. But a proper statistical convergence for the ensemble, represented in terms of a bundle of trajectories, is still unsatisfactory in terms of comparisons with experiments. Can we learn something by these HPC experiments? In this contribution, we show one example, where the occurrence of a chemical reaction in a disordered system is probed. The complex of the copper ion and a segment of the amyloid-β peptide, of wide interest in understanding the progress of Alzheimer's disease, was modelled combining constructions based on empirical force fields with first-principles MD simulations. We simulate a bundle of 16 different structures, biasing different Cu coordination numbers and changing the charge (oxidation state) of the assembly. Even within the given approximations for forces and the poor sampling, we could identify the structures of the complex that are able to react with hydrogen peroxide. The observation explains, at a molecular level, one important linkage between Alzheimer's disease and oxidative stress. This is an example of a general strategy for exploiting reactive configurations within a large set of possible reasonable candidates.

Keywords::

• amyloid peptides
• molecular dynamics
• molecular modeling
• Alzheimer's disease
• oxidative stress

Acknowledgements

The PRACE project pra063, the CALMIP (Calcul en Midi-Pyrénées) project P1049 and the bilateral CNR/CNRS exchange program ‘Identification of active redox pathways for copper ions bonded to amyloid peptides’ are greatly acknowledged for computational resources and funding provided.