Organic fluorescent theranostic molecules have been extensively applied in fields such as biomedical imaging and tumor therapy, attributed to their remarkable features including high sensitivity, excellent biocompatibility, low toxicity, and integrated diagnosis - treatment capabilities. However, as the demand for precision theranostics continues to escalate, traditional molecular design strategies—limited by time-consuming trial-and-error experimentation and prohibitive computational costs—have become insufficient to meet contemporary requirements. Machine learning (ML) has emerged as a revolutionary approach in fluorescent molecule design by directly establishing structure-property relationships, thereby improving the accuracy of theranostic functions and drastically reducing the design cycle. This review systematically collates and analyzes the fluorescent molecule design models based on diverse ML algorithms. It categorizes and comprehensively summarizes the current research by focusing on multiple characteristic attributes of theranostic molecules. Moreover, it puts forward the future development directions of molecular design based on ML methods.

Key words: Machine learning, Fluorescent theranostic molecules, Molecular design, Structure-activity relationship, Molecular descriptors, Database.