2026 Volume 41 Issue 2 Pages 143-149
Drug discovery technologies leveraging machine learning (ML) and generative artificial intelligence (AI) have advanced rapidly in recent years. In the field of targeted protein degraders, such as proteolysis-targeting chimeras (PROTACs), ML- and AI-driven approaches have also attracted increasing attention. To facilitate the optimization of physicochemical and pharmacokinetic properties, including the degradation activity against proteins of interest (POIs) and cell membrane permeability, several ML-based prediction methods have been developed. Moreover, the optimization of linker structures, which are components of PROTACs, is critical for controlling these properties in PROTAC development. However, conventional linker optimization often relies on trial and error based on the intuition and experience of medicinal chemists, resulting in substantial time and labor requirements. To address this, PROTAC linker design methods using various molecular generative AI techniques have been developed. In this article, we review ML- and generative AI-based approaches for PROTAC development, with a particular focus on linker design. In addition, we introduce the ChemTS series, a group of molecular generative AI developed by our research group, and PROTAC-TS, a generative AI-based PROTAC linker design method based on reinforcement learning. We demonstrate the performance of PROTAC-TS by designing PROTAC linkers for three POI ligand–E3 ligand pairs and report the corresponding results. We anticipate that continued accumulation of experimental data, together with further advances in ML technologies, will enable more rational and efficient data-driven strategies for PROTAC design.
