Machine learning

Machine learning methods have been the focus of intense research in recent years, with applications in fields as diverse as chemistry, physics, biology, materials science and engineering. For instance, machine learning methods have been instrumental in our understanding of the relation between chemical structure and functionality, as well as in the engineering of new materials with improved properties. Recent developments have also led to the implementation of machine learning approaches in molecular simulation methods, with, e.g. the use of machine learning-based force fields or through the training of quantitative structure–property relation model over molecular simulation data. This special issue is a selection of six papers reflecting some of the new and exciting developments in the field.

Gabriele Sosso, Volker Deringer, Stephen R. Elliott and Gábor Czányi discuss how machine learning-based interatomic potentials can further our understanding of the thermal properties of amorphous solids, and present applications of neural network potentials to phase-change materials and of Gaussian approximation potentials to amorphous carbon. Joseph C. R. Thacker, Alex L. Wilson, Zak E. Hughes, Matthew J. Burn, Peter I. Maxwell and Paul L. A. Popelier discuss how the machine learning-based force field FFLUX, relying on a topological energy partitioning method known as Interacting Quantum Method, can be applied to the geometry optimization of a peptide-capped glycine. Shriyaa Mittal and Diwakar Shukla examine how machine learning methods, ranging from dimensionality reduction approaches to the more recently developed transfer learning and reinforcement learning approaches, can be used in molecular dynamics simulations to improve the accuracy and efficiency of protein dynamics studies. Ellen Swann, Baichuan Sun, Deidre Cleland and Amanda Barnard discuss how statistical analysis and machine learning can help predict the properties of nanoscale systems and review the relative advantages of the common data representation, reduction and classification methods applicable to molecular and materials modelling. Johannes Hachmann, Mohammad Atif Faiz Afzal, Mojtaba Haghighatlari and Pal Yudhajit report on their recent work on the creation of a software ecosystem that brings together physics-based modelling, high-throughput in silico screening and data analytics, i.e. the use of machine learning and informatics for the validation, mining and modelling of chemical data. Bryce Thurston and Andrew Ferguson discuss how they train a quantitative structure–property relationship (QSPR) model over molecular simulation data of self-assembling synthetic oligopeptides, perform high-throughput screening of oligopeptide chemical space and identify promising candidates for the spontaneous assembly of ordered nanoaggregates with engineered electronic and optical functionality.

I would like to thank the reviewers for their invaluable contribution. I would also like to thank Nick Quirke, Editor-in-Chief of Molecular Simulation, and Jeff Lim, Production Editor, for helping us prepare this special issue.

Jerome Delhommelle
Department of Chemistry, University of North Dakota, Grand Forks, ND, USA
[email protected]; [email protected]