 The proposed AI model, RetroExplainer, automates retrosynthetic research in digital laboratories by formulating the task into a molecular assembly process that utilizes deep learning to guide several retrosynthetic actions. The model's effectiveness is demonstrated on 12 large-scale benchmark datasets and its interpretability allows for transparent decision-making and quantitative attribution. RetroExplainer can also identify multi-step pathways with high accuracy, providing valuable insights for organic synthesis in drug development. This article was authored by Yu Wang, Chao Pang, Yu Xia Wang, and others.