Haomin Wu, Yaxin Xu, Zhiwei Nie, Zhongyang Zhou, Zhengyu Ma, Zhixiang Ren
Induced proximity is changing drug discovery by turning molecular association into a therapeutic design principle. Rather than blocking a single active site, proximity-based agents can degrade, stabilize, relocalize or rewire disease-relevant proteins. This review examines how artificial intelligence is beginning to support that transition, from prioritizing target-effector pairs and modelling ternary or neo-interface assemblies to designing degraders, molecular glues and programmable systems and learning from experimental feedback. We discuss progress across degradative and nondegradative modalities and argue that future impact will depend less on larger models alone than on mechanism-resolved data, realistic benchmarks, interpretable failure analysis and developability-aware design.