Machine learning of correlated dihedral potentials for atomistic molecular force fields

Abstract

Computer simulation increasingly complements experimental efforts to describe nanoscale structure formation. Molecular mechanics simulations and related computational methods fundamentally rely on the accuracy of classical atomistic force fields for the evaluation of inter- and intramolecular energies. One indispensable component of such force fields, in particular for large organic molecules, is the accuracy of molecule-specific dihedral potentials which are the key determinants of molecular flexibility. We show in this work that non-local correlations of dihedral potentials play a decisive role in the description of the total molecular energy—an effect which is neglected in most state-of-the-art dihedral force fields. We furthermore present an efficient machine learning approach to compute intramolecular conformational energies. We demonstrate with the example of α-NPD, a molecule frequently used in organic electronics, that this approach outperforms traditional force fields by decreasing the mean absolute deviations by one order of magnitude to values smaller than 0.37 kcal/mol (16.0 meV) per dihedral angle.

Toward Design of Novel Materials for Organic Electronics

“…We note that the accuracy of the energy disorder not only depends on the quantum mechanical method used in the electronic structure evaluation but also on the methods used to model the embedding, [57] the morphology generation, the underlying force fields, and their parameterization (see Figure 3d about different parameters and their influence on the charge carrier mobility amorphous organic semiconductors), [78] as well as on the consideration of dynamic effects in the model. Due to this strong dependence, small errors in the calculation of the energy disorder lead to large errors in the predicted charge carrier mobility, demanding high accuracy of all models involved in the simulation workflow.…”

Machine learning for accelerating the discovery of high-performance donor/acceptor pairs in non-fullerene organic solar cells

“…Thanks to the contributions of many computer scientists in terms of the combination of artificial intelligence (AI) and big data 22 , machine learning (ML) technique, which is a computational method, spread quickly into many research fields and commercial projects, such as drug discovery 23 , genome sequencing 24 , metalorganic frameworks materials 25 , and organic light-emitting diodes 26 , etc. In material science, ML is an ideal tool that can effectively learn from past massive data sets and mechanisms, automatically generate structures, assess their electronic features and other properties, determine the underlying rules among these data sets and build scientific models to make predictions with reasonable accuracy [27][28][29][30][31][32] . As a result, compared with hundreds of hours or more it takes to evaluate a compound by experiment, the models based on appropriate algorithms can predict properties in a few minutes.…”

Capturing non-local through-bond effects in molecular mechanics force fields I: Fragmenting molecules for quantum chemical torsion scans [Article v1.1]

“…This simple Fourier expansion of a single torsion angle is only an approximation to the true Born-Oppenheimer surface; neighboring torsions can have correlated conformational preferences the low-order Fourier series does not capture [11]. To account for this, 2D spline fits, such as the CMAP potential [12,13], have become a popular way to model non-local correlations by fitting residuals between a 2D QC torsion energy profile of the coupled torsions of interest and the 2D MM torsion energy profile.…”