Deep learning-driven decoding of ubiquitination: from regulatory mechanisms to targeted protein degradation.

backgroundUbiquitination is a highly dynamic post-translational modification that plays central roles in protein homeostasis, signal transduction, immune regulation, and cell fate control. Through the coordinated actions of E3 ubiquitin ligases and deubiquitinating enzymes, ubiquitination shapes the ubiquitin-proteasome system and influences a wide range of physiological and pathological processes. Dysregulation of this system is closely associated with cancer, neurodegenerative disorders, immune dysfunction, and metabolic disease. However, a comprehensive understanding of ubiquitination remains limited because of incomplete annotation, context-dependent regulation, transient molecular interactions, and highly complex many-to-many regulatory networks. MAIN BODY: In this review, we summarize how deep learning is reshaping ubiquitination research at multiple levels. First, we outline the major deep learning architectures applied in this field, including convolutional neural networks, recurrent neural networks, Transformers, protein language models, graph neural networks, generative models, and reinforcement learning. Second, we review recent progress in predicting ubiquitination sites and ubiquitin chain-related features, with emphasis on sequence-based, structure-informed, and multimodal representation learning strategies. Third, we discuss how deep learning contributes to mechanistic decoding of ubiquitination specificity, including substrate recognition by E3 ubiquitin ligases and deubiquitinating enzymes, degron identification, and the organization of ubiquitination regulatory networks. Fourth, we highlight the translational relevance of these approaches in biomarker discovery, targeted protein degradation, molecular glue discovery, degrader optimization, and ubiquitin-centered therapeutic design. Collectively, these advances show that deep learning is not only improving predictive accuracy, but also enhancing mechanistic interpretability and enabling rational molecular design.

conclusionDeep learning is driving ubiquitination research from descriptive prediction toward mechanistic understanding and therapeutic application. Future progress will depend on developing more interpretable models, integrating physiological context, and strengthening experimental validation. Such efforts will accelerate the translation of ubiquitin biology into clinically useful biomarkers, precision diagnostics, and targeted therapies.