Curating Benzothiophene Experimental Absorption and Emission Spectra to Design Fluorescent Organic Polymer Chemical Space: A Machine Learning Quest.

In this study, an approach to design new fluorescent organic polymers based on benzodithiophene (BDT) chromophores are presented by utilizing machine learning (ML) techniques. For this, the BDT chromophores, from the literature, along with their corresponding λ_e. by using Rapid Discovery Kit (RDKit), their molecular descriptors are designed to employ ML models for predicting their λ_max and λ_e properties. Among the evaluated models, Linear Regression, Random Forest and Decision Tree models demonstrate the best performance, achieving R² values between 0.96 and 0.98. Their analysis of SHapley Additive exPlanations (SHAP) values reveals that the Labute Accessible Surface Area (ASA) and the number of Rotatable Bonds can be the most influential features to impact their performance. Leveraging these insights, their 5,000 new polymers are designed with their predicted λ_e extending up to 987 nm. Their highest Synthetic Accessibility Likelihood Index (SALI) scores for the top 1,000 polymers reaches up to 3.21 to indicate their accessibility for synthesis. This work not only advances the understanding of BDT -based materials but can also provide a framework for designing of new fluorescent polymers.