Temperature-transferable coarse-graining of ionic liquids with dual graph convolutional neural networks

Abstract

Computer simulations can provide mechanistic insight into ionic liquids (ILs) and predict the properties of experimentally unrealized ion combinations. However, ILs suffer from a particularly large disparity in the time scales of atomistic and ensemble motion. Coarse-grained models are therefore used in place of costly all-atom simulations, accessing longer time scales and larger systems. Nevertheless, constructing the many-body potential of mean force that defines the structure and dynamics of a coarse-grained system can be complicated and computationally intensive. Machine learning shows great promise for the linked challenges of dimensionality reduction and learning the potential of mean force. To improve the coarse-graining of ILs, we present a neural network model trained on all-atom classical molecular dynamics simulations. The potential of mean force is expressed as two jointly trained neural network interatomic potentials that learn the coupled short-range and many-body long range molecular interactions. These interatomic potentials treat temperature as an explicit input variable to capture its influence on the potential of mean force. The model reproduces structural quantities with high fidelity, outperforms the temperature-independent baseline at capturing dynamics, generalizes to unseen temperatures, and incurs low simulation cost.

Multi-body effects in a coarse-grained protein force field

“…On the other hand, if a slow convergence of the multi-body expansion also holds for other CG protein models, it makes it more challenging to obtain explicit analytical expressions for their effective energy functions. In principle, we expect that, when extended to the recently proposed transferable neural network architecture for the design of CG models, 62,63 multi-body expansion could disentangle the different contributions of interactions between different groups of residues and analytical expression could then be The Journal of Chemical Physics ARTICLE scitation.org/journal/jcp considered (e.g., by means of permutationally invariant polynomials 95 ). However, in addition to the slow convergence of the multibody expansion, the large number of combinations for the different residues' clusters makes this task much more daunting than what has been possible for the characterization of bulk water PES.…”

“…In recent work, by our group 61,62 and others, 26,63,64 a different philosophy has been employed to take into account multibody effects in CG modeling: namely, taking advantage of the ability of modern machine learning techniques to approximate arbitrary complex multi-body functions. Given the recent success in the use of these techniques for the definition of the classical energy function from quantum mechanical calculations, [65][66][67][68][69][70][71][72][73][74][75][76][77][78][79][80][81][82][83][84] a similar idea has been applied for coarse-graining.…”

A review on machine learning algorithms for the ionic liquid chemical space

“…There are few published works on graph-based frameworks for encoding chemical structures for ILs, however these works tend to focus solely on one family of anions or cations and therefore their extension and generalisation might still be limited. [66][67][68] Another family of descriptors used in QSPR methods are those of quantum chemical (QC) or thermodynamic nature. QC descriptors use values from quantum calculations, such as HOMO and LUMO energies, polarity, electron affinity, electronegativity etc.…”

Dynamical properties across different coarse-grained models for ionic liquids

“…Regarding dynamics, it was found that the diffusive speedup with respect to the AA model differs for the cation and anions. We also note the development of other CG models for RTILs using alternative methodologies: Newton inversion [40] (also known as inverse Monte Carlo), relative entropy [41], and graph neural networks [42]. While each of these studies features different parameterization schemes and a variety of resulting properties, they share the common difficulty of representability-simultaneous reproduction of structural, thermodynamic, and dynamical properties.…”