O-GlcPRED-Ens: An Ensemble Deep Learning Approach for Predicting O-linked Glycosylation Sites in Proteins

Introduction: O-linked Glycosylation is a significant post-translational modification in which N-acetylgalactosamine (N-GalNAc) is added to the hydroxyl group of serine (S) or threonine (T) residues. This modification plays a crucial role in regulating protein function and cell signaling. However, not all serine or threonine sites serve as O-linked Glycosylation sites, making the accurate identification of these sites in protein sequences essential for understanding their biological roles. This study aims to develop an effective predictive model for identifying O-linked Glycosylation sites within protein sequences. Methods: An integrated model named O-GlcPRED-Ens was developed, which combines a Time Distributed layer, Convolutional Neural Network (CNN), and Bidirectional Long Short-Term Memory (BiLSTM) architecture. This model utilizes three types of sequence information: contextual embeddings generated by the ESM-2 protein language model, features from AAIndex, and representations from Word2Vec. We constructed two network frameworks and optimized systematic parameters to enhance model performance. Results: An ensemble learning-based predictive model, O-GlcPRED-Ens, was developed to identify O-glycosylation sites by integrating various feature encoding techniques, including ESM-2, AAIndex, and Word2Vec. Discussion: Extensive comparative analyses were conducted using human and mouse test sets, along with independent test sets from other species. The results demonstrate a significant increase in both sensitivity and accuracy for predicting O-linked Glycosylation sites, indicating that OGlcPRED-Ens outperforms existing prediction tools. Conclusion: The O-GlcPRED-Ens model represents a significant advancement in predicting Olinked Glycosylation sites, thereby enhancing our ability to identify these critical protein modifications. This improvement accelerates the discovery of unknown O-linked Glycosylation sites, thereby deepening our understanding of their biological significance and potential roles in cellular processes