• Comprehensive review of computational approaches in PROTAC drug discovery • Examines challenges in ternary complex modelling and cooperativity prediction • Integrates molecular dynamics, free energy, and AI-driven linker optimization • Discusses machine learning and systems biology frameworks for degrader design • Outlines translational opportunities and computational frontiers in targeted degradation Proteolysis-targeting chimaeras (PROTACs) represent a transformative therapeutic strategy that exploits the ubiquitin–proteasome system to selectively degrade disease-relevant proteins. By shifting pharmacology from occupancy-driven inhibition to event-driven degradation, PROTACs have expanded the range of druggable targets to include transcription factors, scaffolding proteins, and mutation-prone oncogenic drivers. However, rational design of degraders remains computationally challenging because their activity depends on the formation of dynamic ternary complexes between the protein of interest, the PROTAC molecule, and an E3 ligase. This review examines the computational foundations of PROTAC discovery. We first discuss the mechanistic determinants of ternary complex formation, including cooperativity (α) and its relationship to degradation efficiency metrics such as DC50 and Dmax. We then survey the evolution of computational approaches used in degrader design, beginning with binary docking and pharmacophore modelling and progressing to specialised ternary complex modelling platforms such as PRosettaC, PROTAC-Model, and DegraderTCM. The role of molecular dynamics simulations and enhanced sampling methods in capturing linker flexibility, ternary stability, and ubiquitination-competent conformations is examined, alongside free-energy approaches (MM-GBSA, FEP, PMF) used to quantify cooperativity and guide chemical optimisation. We further discuss computational strategies for linker design, including conformational sampling approaches and emerging generative models such as DiffLinker and reinforcement-learning pipelines. Advances in machine learning, including graph neural networks for ternary complex prediction and PROTAC-specific ADMET models, are reviewed together with hybrid AI–physics workflows that combine data-driven and simulation-based design. Finally, we consider the expanding repertoire of recruitable E3 ligases beyond CRBN and VHL, and discuss how transcriptomic resources such as GTEx and single-cell RNA sequencing datasets can inform tissue-selective degrader design. We conclude by outlining key challenges, including limited ternary structural data, force-field limitations for large flexible molecules, and the absence of standardised benchmarking datasets, and highlight future opportunities arising from the integration of cryo-EM structural biology, AlphaFold-based modelling, and AI-guided computational platforms for next-generation PROTAC discovery.
Tewari et al. (Fri,) studied this question.