Large Language Model-Assisted Prediction and Virtual Screening of Near-Infrared-II Fluorescent Molecules

Fluorescence imaging in the second near-infrared window (NIR-II, 1000-1700 nm) has attracted increasing attention in biomedical applications owing to its deep tissue penetration, high spatial resolution, and low background autofluorescence. However, the rational optimization of high-performance NIR-II fluorophores remains challenging because of the complex relationship between molecular structure and optical properties. In this study, we integrated existing fluorescence datasets with NIR-II fluorophore data to construct a comprehensive dataset covering maximum emission wavelengths from the visible region to the NIR-II window. Based on this dataset, we developed a large language model (LLM)-based framework for predicting molecular maximum emission wavelengths and compared its performance with conventional machine learning and deep learning models. The results demonstrated that the LLM achieved competitive predictive performance compared with the baseline models, highlighting the feasibility of LLM-based structure-property prediction for fluorescent molecules. Model-guided virtual screening was subsequently performed to prioritize candidate NIR-II fluorophores, leading to the identification of two promising candidates that were experimentally characterized and exhibited favorable imaging performance in vitro and in vivo. Overall, this study demonstrates the feasibility of LLM-assisted structure-property prediction and virtual screening for NIR-II fluorophores, providing an efficient computational strategy for candidate prioritization and molecular optimization.