Application of Machine Learning in Developing Quantitative Structure–Property Relationship for Electronic Properties of Polyaromatic Compounds

Abstract

The degree of π orbital overlap (DPO) model has been demonstrated to be an excellent quantitative structure–property relationship (QSPR) that can map two-dimensional structural information of polycyclic aromatic hydrocarbons (PAHs) and thienoacenes to their electronic properties, namely, band gaps, electron affinities, and ionization potentials. However, the model suffers from significant limitations that narrow its applications due to inefficient manual procedures in parameter optimization and descriptor formulation. In this work, we developed a machine learning (ML)-based method for efficiently optimizing DPO parameters and proposed a truncated DPO descriptor, which is simple enough that can be automatically extracted from simplified molecular-input line-entry system strings of PAHs and thienoacenes. Compared with the result from our previous studies, the ML-based methodology can optimize DPO parameters with four times fewer data, while it can achieve the same level of accuracy in predictions of the mentioned electronic properties to within 0.1 eV. The truncated DPO model also has similar accuracy to the full DPO model. Consequently, the ML-based DPO approach coupled with the truncated DPO model enables new possibilities for developing automatic pipelines for high-throughput screening and investigating new QSPR for new chemical classes.

Machine Learning-Based Quantitative Structure–Property Relationships for the Electronic Properties of Cyano Polycyclic Aromatic Hydrocarbons

“…In this study, the truncated DPO model is optimized with the ML-based method, which is an iterative process that has been proposed previously for optimizing the DPO model’s parameters. To concisely include the new S DPO descriptor in this process, let X i [ t ] be a vector that is composed of both the DPO value and the S DPO value of the i -th compound as below: X i false[ t false] = [ lefttrue normalD normalP normalO i ( a [ t ] , b [ t ] , c [ t ] , d [ t ] ) S D P O , i ( s [ t ] ) 0.25em ] where the superscript t denotes the t th iteration.…”

“…Consequently, errors of the bandgap and EA, IP properties converge to around 0.12 and 0.10–0.11eV, respectively, with only 50 training data points, roughly 10% of the training set. This finding is consistent with our previous study for PAH and thienoacene chemical classes.…”

“…The application of the DPO model to the thienoacene chemical class 9 illustrated how the DPO model treats aromatic heteroatomic rings fused to PAH rings. This study, combined with the ML-DPO methodology, 11 It is found that the errors of the models are all less than 0.20 eV with average errors of around 0.12 and 0.10 eV for bandgap and EA, IP, respectively. Furthermore, the model converges and can achieve this level of error with only 50 training data points, which is quite small in comparison to the total data set.…”

“…Parameters Optimizing with the ML-Based DPO Model. In this study, the truncated DPO model is optimized with the ML-based method, 11 which is an iterative process that has been proposed previously for optimizing the DPO model's parameters. To concisely include the new S DPO descriptor in this process, let X i…”

“…In particular, the training set with 556 molecules approximates 60% of the total, and the test set includes 370 molecules. A random yet stratified splitting procedure described in our previous work was used to construct the training set to preserve the distribution of bandgap values of the original data set . To achieve this, the total data set is divided into bins with a bandgap width of 0.5 eV.…”