Exploring New Azobenzene Type Photoswitches by Machine Learning with Lowest Possible Fluorescence Excitons with Ease of Synthesis.

Current investigation presents the design and analysis of 299 azobenzene photoswitches (PSs) for their lowest possible π→ π* transition energies along with their predicted emission maxima values through machine learning (ML) analysis. Their π→ π* transitions related wavelength is calculated by Erying equation to reveal its range up to 256 nm. Their Synthetic Accessibility Likelihood Index (SALI) indicates that a substantial number of them can be synthesized with ease. Among various tested ML model, eXtream Gradient Boosting (XGBoost) regression models demonstrates its high accuracy by achieving an R² value of 0.87. Their designed molecular descriptors show its Maximum Electrotopological State Index (MaxEStateIndex) to impact the model most. For its emission wavelengths, the random forest regression model yields its promising results with its R² of 0.92 and a Mean Squared Error (MSE) of 0.38. Its SHAP value reveals the top contributing descriptors being Estate_VSA5, NumValenceElectrons, Estate_VSA3, Chi0n, Chi1v, PEOE_VSA9, Chi0v, and VSA_Estate2. This work not only expands the library of azobenzene PSs but also enhances their understanding of their electronic properties for their future applications in materials science.