Abstract
Molecular generative models (MGMs) are rapidly reshaping the landscape of drug discovery. To date, thousands of MGMs have been reported, some of which facilitate the design and optimization of molecules with the desired biological activities and pharmacological properties. In the context of drug discovery, we classify common MGMs into three primary categories: ligand-based molecular generation (LBMG), structure-based molecular generation (SBMG), and pharmacophore-based molecular generation (PBMG). This perspective discusses the principles underlying these MGMs, their evolution, and typical examples of LBMG, SBMG, and PBMG, along with their successful applications in drug discovery. Additionally, we discuss their respective advantages, limitations, and potential future research directions with the hope of promoting the continued development and application of MGMs in advancing drug discovery.