CoNglyPred: Accurate Prediction of N‐Linked Glycosylation Sites Using ESM‐2 and Structural Features With Graph Network and Co‐Attention

• DOI: 10.1002/pmic.202400210
• Publication Date: 2024-10-03T13:50:16Z
• Authors: Hongmei Wang; Long Zhao; Ziyuan Yu; Ximin Zeng; Shaoping Shi

ABSTRACT N‐Linked glycosylation is crucial for various biological processes such as protein folding, immune response, and cellular transport. Traditional experimental methods determining N‐linked sites entail substantial time labor investment, which has led to the development of computational approaches a more efficient alternative. However, due limited availability 3D structural data, existing prediction often struggle fully utilize information fall short in integrating sequence effectively. Motivated by progress pretrained language models (pLMs) breakthrough structure prediction, we introduced high‐accuracy model called CoNglyPred. Having compared pLMs, opt large‐scale pLM ESM‐2 extract embeddings, thus mitigating certain limitations associated with manual feature extraction. Meanwhile, our approach employs graph transformer network process structures predicted AlphaFold2. The final output embedding are intricately integrated through co‐attention mechanism. Among series comprehensive experiments on independent test dataset, CoNglyPred outperforms state‐of‐the‐art demonstrates exceptional performance case study. In addition, first report uncertainty predictors using expected calibration error error.