Abstract: The optimization of drug molecular structure is a crucial step in enhancing the efficacy and safety of drugs. Its primary objective is to rationally adjust the molecular structure to meet the strict requirements for the comprehensive performance of drugs during the research and development process. However, the vast chemical space of drugs and the complex multi-objective optimization tasks pose significant challenges to this process. In recent years, machine learning techniques, represented by deep learning and reinforcement learning, have been widely applied in the optimization of drug molecular structures due to their powerful capabilities in information mining, feature extraction, and nonlinear fitting, providing new ideas and methods for drug design. This paper systematically reviews the latest research progress of machine learning in the optimization of drug molecular structures, deeply analyzes and compares their application advantages and characteristics in different structural optimization scenarios, summarizes the main challenges currently faced in the optimization of drug molecular structures, and provides a technical perspective on the possible research directions for future structural optimization.