Site-specific prediction of O-GlcNAc modification in proteins using evolutionary scale model

Area of Science:

• Biochemistry
• Computational Biology
• Proteomics

Background:

• Protein glycosylation, specifically O-GlcNAcylation, is a crucial post-translational modification impacting biological processes and disease.
• Computational methods, including machine learning and protein language models, are increasingly used for predicting O-GlcNAc sites, offering efficiency and cost reduction.

Purpose of the Study:

• To evaluate the efficacy of the Evolutionary Scale Model 2 (ESM-2) for predicting O-GlcNAc glycosylation sites in human proteins.
• To establish a computational approach for O-GlcNAc site prediction using ESM-2, addressing a gap in existing literature.

Main Methods:

• Utilized approximately 1100 O-linked glycoprotein sequences from the O-GlcNAc database for model training.
• Employed the ESM-2 model, a protein language model, for predicting O-GlcNAc sites in human proteins.
• Compared ESM-2 performance against traditional models to assess accuracy and overfitting tendencies.

Main Results:

• The ESM-2 model showed consistent improvement during training, achieving an accuracy of 78.30%, recall of 78.30%, precision of 61.31%, and F1-score of 68.74%.
• ESM-2 demonstrated superior performance over traditional models by exhibiting optimal training and testing predictions and avoiding significant overfitting.
• The model's effectiveness in predicting O-GlcNAc sites in human proteins was validated.

Conclusions:

• The ESM-2 model is effective for accurate O-GlcNAc site prediction in human proteins.
• Accurate O-GlcNAc site prediction can significantly advance glycoproteomic research, aiding in understanding protein function, disease mechanisms, and therapeutic development.
• Future research should explore diverse data, longer sequences, and enhanced computational resources to further refine prediction models.