Abstract: Artificial intelligence (AI) algorithm has attracted an increasing amount of attention and achieved some amazing progress in drug design. Molecular generation as a unique applied technique demonstrates the potential to replace traditional drug design by medicinal chemists. To better utilize it in solving real problems, it is essential to have an accurate molecular property prediction model, a 3D structure generation technique and a unified benchmark. This review summarizes recent advances in the application of AI algorithms in the design of novel drug structures, with particular focus on the technical details as well as the advantages and disadvantages of different molecular representation forms and neural network architectures.