Interpretable Machine Learning Combined TD-DFT Calculations for the Study of Colorless Transparency Polyimides
Xu Li, Haoyu Yang, Yonghong Tao, Qingji Wang, Tongfei Shi, Lin Li
Abstract
The application of polyimides (PIs) is extensive in flexible electronics, artificial intelligence, and chip technology. However, the relationship between the structure of PIs and their colorless transparency remains unclear. In this study, we established a data set to investigate the colorless transparency of PIs by combining machine learning (ML) with time-dependent density functional theory (TD-DFT) calculations. The quantitative structure–activity relationship (QSAR) models were constructed to represent the colorless transparency of PIs, using 46 molecular descriptors derived from TD-DFT calculations. Comparing various ML algorithms, gradient boosting (GB), and support vector regression (SVR) models exhibited optimal performance for the colorlessness and transparency of PIs, respectively. Pearson correlation coefficient and SHAP model were employed to illustrate the role of significant descriptors in determining the colorlessness and transparency of PIs. Furthermore, the key substructures serving as ten diamine units were identified based on the data-driven analysis of the target properties of PIs. The research approach will provide theoretical guidance for the targeted synthesis of colorless and transparent PIs and introduce an innovative concept for the design of novel materials.