Machine Learning Approaches for Determining Molecular Packing of Organic Semiconductors: Toward Accurate Crystal Structure Prediction

Organic semiconductors (OSCs) with π-electron skeletons (π-cores) have attracted much attention. The creation of more innovative new molecules with high carrier mobility requires the strategic molecular design. One of the most important properties affecting the carrier mobility of π-conjugated OSCs is the molecular arrangement, especially the two-dimensional (2D) molecular packing of the π-cores which exhibit 1D or 2D carrier conductivity as follows: π-stacking, herringbone (HB) packing, and brickwork (BW). Since two molecular packing structures, HB and BW, have not been predicted theoretically, chemists have typically designed new OSC molecules based on their previous knowledge so far. Therefore, the use of computational science and informatics science is eagerly needed to strategically design an OSC molecule with unprecedented properties and functions. In this study, we introduce machine learning method to determine whether an OSC forms 2D HB or BW molecular packing with a high accuracy. We further present the computational method to predict the crystal structure of an OSC only from its chemical structure using molecular mechanics calculation and molecular dynamics simulation coupled with our proposed machine learning model to classify the type of 2D molecular packing. The combination of molecular simulations with the machine learning model has the potential to predict the crystal structure of organic molecules.