NTpred

NTpred predicts tyrosine nitration (NT) sites in protein sequences to identify post-translational modification positions implicated in disease-related protein dysfunction.

Key Features:

• Target PTM: Targets tyrosine nitration (NT) as the post-translational modification of interest in protein sequences.
• Sequence encoding: Extracts features from raw protein sequences using four distinct sequence encoders.
• Feature spaces: Constructs eight unique feature spaces, including fused combinations of individual encodings.
• Feature selection (RFE): Applies Recursive Feature Elimination (RFE) across feature spaces to remove irrelevant or redundant features.
• Base classifiers: Trains eight Gradient Boosted Tree classifiers on selected features from individual encodings and fusion vectors.
• Probabilistic meta-feature: Generates a probabilistic feature space from the probability scores of the eight Gradient Boosted Tree classifiers.
• Final classifier: Trains a Logistic Regression classifier on the probabilistic feature space as the final predictive model.
• Benchmark performance: Shows improved performance over existing predictors on BD1 (MCC +13.7%, AUC +20.1%) and BD2 (MCC +5.3%, AUC +1.0%).

Scientific Applications:

• Disease mechanism analysis: Supports analysis of protein nitration roles in neurodegenerative disorders, cardiovascular conditions, autoimmune diseases, and carcinogenesis.
• Drug discovery: Aids identification of nitration sites to inform therapeutic strategies and drug discovery targeting protein modification pathways.

Methodology:

Extract features from raw protein sequences using four sequence encoders to form eight feature spaces (including fused encodings); apply Recursive Feature Elimination (RFE) for feature selection; train eight Gradient Boosted Tree classifiers on selected features; construct a probabilistic feature space from classifier probability scores; and train a Logistic Regression classifier as the final predictive model.