Rational design of PROTACs empowered by deep generative modeling: current advances and future perspectives

Deep generative modeling (DGM) represents a transformative paradigm in proteolysis-targeting chimera (PROTAC) discovery, transitioning the field from empirical trial-and-error and computer-aided drug design (CADD) toward data-driven, autonomous candidate generation. PROTAC design presents a high-dimensional systems engineering challenge due to the intertwined complexities of ternary-complex cooperativity, linker conformational plasticity, and conflicting pharmacokinetic profiles. By systematic integration of machine learning architectures-spanning language models, graph neural networks, and diffusion frameworks-with target-aware optimization strategies, modern workflows facilitate the rational design of linkers, warheads, and tripartite chimeras. While data sparsity and multi-objective optimization remain key computational bottlenecks, closed-loop simulation pipelines and experimental validation are overcoming these challenges. We systematically review state-of-the-art DGM methodologies for de novo PROTAC generation and delineate future strategies for harnessing these generative models to produce clinically translatable degraders.