Graphogly: a Protein Post-Translational Modification Classification Model Based on Contextual Protein Language Embedding and Graph Convolutional Neural Networks

O-linked glycosylation is a complex form of posttranslational modification in human proteins that plays a crucial regulatory role in a wide range of cell types. It is intimately involved in cellular metabolic activities and signalling networks. In particular, abnormal patterns of O-linked glycosylation have been implicated in the onset and progression of many diseases, including cancer and diabetes. Therefore, the accurate identification of O-linked threonine glycosylation sites (OTGs) remains a challenging task that usually requires extensive laboratory experimentation. In this study we use the GraphOGly model to identify O-linked threonine glycosylation sites. In the data representation phase, we use ProtT5, a pre-trained protein language model, to generate contextual embeddings for individual amino acid residues. This is followed by the application of a graph neural network architecture consisting of three GCNConv layers for feature extraction, together with a self-attention layer and a multilayer perceptron module for further representation learning. The experimental results demonstrate that GraphOGly is superior to other existing tools in predicting O-linked glycosylation sites, and the final experimental results after 5-fold cross-validation are ACC 88.73%,Sn 88.96%,Sp 88.49%,MCC 0.7748,AUC 0.9408.