Integrating Machine Learning and SHAP Analysis to Advance the Rational Design of Benzothiadiazole Derivatives with Tailored Photophysical Properties

Abstract

2,1,3-Benzothiadiazole (BTD) derivatives show promise in advanced photophysical applications, but designing molecules with optimal desired properties remains challenging due to complex structure–property relationships. Existing computational methods have a high cost when predicting precise photophysical characteristics. Machine learning with Morgan fingerprints was employed to forecast BTD derivative maximum absorption and emission wavelengths. Three flavors of machine learning models were applied, namely, Random Forest, LigthGBM, and XGBoost. Random forest achieved R2values of 0.92 for absorption and 0.89 for emission, validated internally with 10-fold cross-validations and externally with recent experimental data. SHapley Additive exPlanations (SHAP) analysis revealed critical design insights, highlighting the tertiary amine presence and solvent polarity as key drivers of red-shifted emissions. By the development of a web-based predictive tool, the potential of machine learning to accelerate molecular design is demonstrated, providing researchers a powerful approach to engineer BTD derivatives with enhanced photophysical properties.