EMNGly: predicting N-linked glycosylation sites using the language models for feature extraction

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Bioinformatics

Background:

• N-linked glycosylation is a vital post-translational modification impacting protein folding, stability, and function.
• Dysregulation of N-linked glycosylation is linked to various diseases, highlighting the need for accurate site identification.
• Experimental identification of N-linked glycosylation sites is complex, driving the need for computational approaches.

Purpose of the Study:

• To develop and evaluate a novel computational approach for predicting N-linked glycosylation sites.
• To leverage pretrained protein language and structure models for enhanced feature extraction in N-linked glycosylation site prediction.

Main Methods:

• The EMNGly approach was developed, integrating features from pretrained Evolutionary Scale Modeling (ESM) and Inverse Folding Model (IFM).
• A support vector machine (SVM) classifier was employed for predicting N-linked glycosylation sites based on extracted features.
• Rigorous evaluation was performed using ten-fold cross-validation and independent test sets.

Main Results:

• EMNGly demonstrated superior performance compared to existing N-linked glycosylation site prediction methods.
• The approach achieved high predictive accuracy, with a Matthews Correlation Coefficient (MCC) of 0.8282.
• Exceptional performance metrics were recorded on an independent test set, including sensitivity (0.9343), specificity (0.8934), and accuracy (0.9143).

Conclusions:

• EMNGly represents a significant advancement in computational N-linked glycosylation site prediction.
• The integration of advanced protein language and structure models enhances prediction accuracy.
• This method provides a valuable tool for researchers studying protein modification and glycosylation-related diseases.