Customized Carbon Dots with Predictable Optical Properties Synthesized at Room Temperature Guided by Machine Learning

Fluorescent carbon dots (CDs) have been increasingly used in fluorescence detection and imaging based on their tunable fluorescence (FL) and resistance to photobleaching. However, the fast and reliable design of fluorescent CDs with specific optical properties involves a number of factors, such as the concentration of precursors, reaction time, and solvents. Therefore, it is usually considered difficult to design CDs with favorable optical properties. Herein, we report an extreme gradient boosting (XGBoost) model for guiding the fabrication of CDs with high FL intensity and tunable emission from p-benzoquinone (PBQ) and ethylenediamine (EDA) in different solvents at room temperature. Among a variety of studied machine learning models, XGBoost shows the best performance in the field of material synthesis, with a prediction coefficient of determination (R2) higher than 0.96. The XGBoost model can effectively predict the optical properties of CDs, including the maximum FL intensity and emission centers. Guided by the XGBoost model, various green or blue fluorescent CDs with adjustable emission centers and solubility properties are designed and fabricated accurately. These CDs are successfully applied for Fe3+ detection, sustained drug release, whole-cell imaging, and poly(vinyl alcohol) (PVA) film preparation. These results suggest the great potential of the combination of machine learning and CD synthesis as an effective strategy to help researchers realize accurate selection of reasonable CDs with individual customized properties to achieve different goals.