Virtual mining of polymer monomers for photodetectors application and regression-aided reorganization energy prediction

[Display omitted] •Machine learning-based approach is presented for virtual mining of polymer monomers for photodetectors.•DFT computations are used to predict the reorganization energy of >500 organic molecules.•Similarity analysis is applied to screen easily synthesizable polymer monomers for designing high-performance photodetectors.•Numerous regression models are developed and tested/selected their predictive capability.•Library enumeration and fitness score strategies are used.•SYnthetic BAyesian classifier (SYBA) is used to test the synthetic accessibility of designed monomers. Predicting and understanding the charge transport properties of organic semiconductors is a crucial target for constructing efficient electronic devices such as photodetectors. To this end, reorganization energy (Re) is a key parameter for molecular design, which provides quantitative strength of electron-photon coupling process (charge transport parameter). Although modern density functional theory-based approaches can accurately simulate intramolecular RE trends, such computations are time-consuming and costly. In this study, we present machine-learning tools for the accurate and fast prediction of intramolecular Re. These tools are effectively used for screening of new compounds with low internal Re. Moreover, virtual mining of polymer monomers is performed for photodetector applications. Machine learning models are developed with molecular descriptors (features). Regression analysis is used to show how two or more variables are related. The dependent variable yields the best outcomes and helps to predict Re. This research shows that chemical designs can accurately predict the REs of organic semiconductor materials. Out of five different regressions models, gradient booster regression gives the best prediction capability (R2 = 0.761). Moreover, chemical similarity analysis (CSA) is used to search for a new group of molecules with high performance. Library enumeration is used for structure determination and for studying the properties of these small molecules. The fitness score of newly designed quinoxaline is more than 0.9, which shows best outcome. The new strategy holds immense potential for the virtual mining of polymer monomers for photodetectors applications and regression-aided reorganization energy predictions.