From undruggable to degradable: A deep learning-enabled framework for precision orthopaedic protein degradation

A fundamental translational impasse in musculoskeletal disorders-from osteoarthritis to bone metastases-is the "undruggability" of critical pathological drivers sequestered within anatomically restricted and avascular skeletal niches. While Targeted Protein Degradation (TPD) offers a revolutionary event-driven modality to eradicate these refractory targets, its clinical translation to orthopaedics is severely hindered by the physicochemical incompatibility of large degraders with dense extracellular matrices and the scarcity of bone-specific E3 ligases.

In this review, we present a transformative framework integrating deep learning (DL) with multimodal omics to circumvent these barriers. We systematically examine how emerging DL architectures-spanning geometric deep learning, protein language models, and generative design-are redefining the orthopaedic TPD pipeline.

We highlight computational strategies for prioritizing cryptic pockets on "undruggable" skeletal transcription factors, identifying tissue-restricted E3 ligases via single-cell transcriptomics to minimize systemic toxicity, and optimizing degrader permeability ("penetrability-first" design) to navigate dense cartilage and bone matrices.

By bridging the gap between computational prediction and skeletal pathobiology, this roadmap outlines a shift in orthopaedic care from palliative symptom management to precise, mechanism-based microenvironmental reprogramming.

This review provides a timely, computationally-driven roadmap to translate Targeted Protein Degradation (TPD) from oncology to orthopaedics. By integrating deep learning with skeletal pathobiology to overcome matrix penetrability and off-target toxicity, this framework accelerates the development of precision degraders. Ultimately, it offers a tangible strategy to shift the clinical management of joint degeneration, osteoporosis, and bone tumors from palliative care to durable, mechanism-based microenvironmental reprogramming.