PROTACs, also called proximity-inducing agents, are chimeric molecules composed of a ligand for protein of interest (POI), an E3 ligase ligand and a linker connecting them. PROTACs have transformed the therapeutic landscape by enabling an event-driven strategy to degrade disease-associated proteins previously regarded as undruggable. The unique event-driven mechanism of PROTACs allows selective protein degradation with greater potency and lower drug resistance than conventional occupancy-based inhibitors. Despite their advantages, challenges such as high molecular weight, low permeability, poor pharmacokinetic properties restrict their clinical applications. To overcome these limitations, AI-driven technologies are being utilised to generate novel, chemically valid PROTACs. This review highlights the drawbacks of conventional computational methods and explores emerging AI-driven tools applied to multiple areas of PROTAC research, such as target (POI) selection (DeepUSI, DrugnomeAI), linker generation (AIMLinker, DiffLinker), activity prediction (AI-DPAPT, DeepPROTAC), POI degradability assessment (PrePROTAC, MAPD), ternary complex modelling (ProFlow), PROTAC generation (PROTAC-RL), and ADME property estimation (MT-GNN). It also outlines current challenges such as data scarcity, reproducibility issues, inadequate model generalizability, emphasizing the need for hybrid models or integrated AI techniques to mitigate these limitations.
Ghosh et al. (2026) studied this question.