Enhanced O-glycosylation site prediction using explainable machine learning technique with spatial local environment.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology
• Genomics and Proteomics

Background:

• Accurate prediction of O-GlcNAcylation sites is vital for understanding disease mechanisms and therapeutic development.
• Existing machine learning models often overlook spatial interactions of amino acids due to reliance on primary/secondary structures.

Purpose of the Study:

• To develop a novel machine learning approach for predicting O-GlcNAcylation sites.
• To incorporate three-dimensional spatial information using local environmental features.
• To enhance prediction accuracy and identify key factors influencing O-GlcNAcylation.

Main Methods:

• Introduction of local environmental features incorporating 3D spatial information.
• Utilization of sparse recurrent neural networks to capture protein sequential nature.
• Development of an explainable machine learning model for identifying key influencing factors.

Main Results:

• The proposed model achieved an F1 score of 28.3%.
• Feature selection using the top 20% of features resulted in the highest F1 score of 32.02%, a 1.4-fold improvement over existing PTM models.
• Statistical analysis confirmed the relevance of the top 20 features with existing literature.

Conclusions:

• The novel approach significantly improves O-GlcNAcylation site prediction accuracy.
• The method provides insights into key factors influencing O-GlcNAcylation, aiding further research.
• The developed model and features pave the way for better understanding and targeting of O-GlcNAcylation in diseases.