DeepO-GlyThr: an interpretable deep learning framework for predicting O-linked threonine glycosites in human proteins

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• O-linked glycosylation is a crucial protein post-translational modification.
• Predicting O-linked threonine glycosites computationally is challenging and lacks interpretability.

Purpose of the Study:

• To develop an interpretable deep learning framework for predicting human O-linked threonine glycosites.
• To improve the accuracy and interpretability of O-linked threonine glycosite prediction.

Main Methods:

• Developed DeepO-GlyThr, a deep learning framework integrating CNN, BiGRU, and attention modules.
• Utilized sequence features and interpretability techniques like Attention and Integrated Gradients.
• Validated performance on an independent test set.

Main Results:

• DeepO-GlyThr achieved high performance metrics: 0.885 sensitivity, 0.900 accuracy, 0.898 F1-score, and 0.800 MCC.
• Outperformed existing methods in O-linked threonine glycosite prediction.
• Interpretability analysis highlighted the importance of local physicochemical and positional features.

Conclusions:

• DeepO-GlyThr offers an accurate and interpretable solution for O-linked threonine glycosite prediction.
• The framework provides insights into sequence patterns influencing glycosite selection.
• A web server is available to facilitate the use of DeepO-GlyThr.