Neighborhood Path Complex for the Quantitative Analysis of the Structure and Stability of Carboranes

Abstract

Thanks to the tremendous progress in data, computing power and algorithms, AI-based material mining and design have gained much attention. However, building high-performance AI models requires efficient material structure representation. In this work, we propose a structural characterization method based on the neighborhood path complex for the first time. Specifically, we use persistent neighborhood path homology to obtain the structural features by introducing a filtration. This approach preserves more elemental information, as well as the corresponding physicochemical information, through the directed edges of the neighborhood digraph. To validate our model, we perform cross-validation with the carborane structures. The Pearson coefficient for stability prediction is as high as 0.903, which is a 15.5% improvement compared to the traditional persistent homology method. In addition, we constructed a prediction model based on the neighborhood path complex, and the Pearson coefficients for the prediction of carboranes' HOMO, LUMO, and HOMO-LUMO gaps were 0.915, 0.946, and 0.941, respectively. The results show that our proposed method can effectively extract structural information and achieve accurate material property prediction.