DeepO-GlcNAc

DeepO-GlcNAc predicts protein O-GlcNAcylation sites using a deep learning framework that combines Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and attention mechanisms to improve prediction performance for this dynamic post-translational modification.

Key Features:

• Deep learning architecture: The framework integrates LSTM, CNN, and attention mechanisms for O-GlcNAcylation site prediction.
• Cross-species robustness: The model demonstrates strong robustness across five cross-species datasets excluding humans.
• Performance evaluation: The tool achieves 92% accuracy, 72% average precision, 0.60 MCC, and 92% AUC in ROC analysis on independent test data.
• Ablation validation: Ablation studies confirm that attention mechanisms and LSTM components contribute positively to prediction performance.

Scientific Applications:

• O-GlcNAc site prediction: Bioinformatic prediction of protein O-GlcNAcylation sites prior to experimental validation in O-GlcNAc research.
• Cross-species analysis: Investigation of O-GlcNAcylation patterns across multiple species using the trained model.
• Comparative benchmarking: Evaluation against external predictors using independent datasets for method validation.

Methodology:

DeepO-GlcNAc was developed using an annotated unbalanced O-GlcNAcylation dataset and a deep learning framework combining LSTM, CNN, and attention mechanism. Model performance was validated through ablation studies, five cross-species datasets, and comparison with three external predictors on an independent dataset.