Crystal system and space group prediction of two-dimensional materials from chemical formula via deep neural networks

Abstract

Predictive modeling of undiscovered two-dimensional (2D) materials plays an important role in 2D materials research. In the initial stage of modeling, predicting the crystal system and space group, two fundamental symmetry parameters of a crystal, from their chemical formula is essential for global structure search and discovery of new 2D crystalline materials. Here, we develop a deep neural network (DNN) model containing 290 descriptors to predict the crystal system and space group of 2D materials in AA stacking based on their chemical formulas. Our model achieves an accuracy of 74.56% in predicting the crystal system and 73.14% in predicting the space group with significant generality. In addition, the probability of obtaining the correct space group from the top five candidates reaches 89.58%. Our results also demonstrate that approximate predictions can be generated with the top 80 descriptors with the predicted results close to those obtained with all 290 descriptors, indicating the relative feature importance difference of descriptors. This study provides an effective DNN model for the quick determination of crystal systems and space groups for the global structure search of 2D materials.