Plug-in Models: A Promising Direction for Molecular Generation
Ningfeng Liu, Hongwei Jin, Liangren Zhang, Zhenming Liu · Health Data Science · 2023
The molecular generation has emerged as a powerful tool for computer-aided drug design in recent years, as it can explore a large and unknown chemical space and discover novel structures or scaffolds.Furthermore, a candidate compound needs to satisfy multiple criteria, such as target affinity, pharmacokinetics, toxicity, synthetic accessibility, etc., to pass clinical trials and meet industrial standards.Therefore, multi-objective methods have become a focal point of molecular generation and optimization.Several reviews have been published recently to summarize previous works in molecular generation and categorize them (Table ).In this article, we propose a classification scheme based on both the model's architecture and its practical use, namely, entrenched or plug-in models, especially for multi-objective molecular generation models.We argue that plug-in methods have superior flexibility in both model building and practical use, broader application potential, and higher performance boundaries, and they deserve more attention in the future.