HydPred

HydPred predicts protein hydroxylation sites (hydroxyproline and hydroxylysine) to identify post-translational modification positions relevant to human disease mechanisms.

Key Features:

• Methodology: Combines the synthetic minority over-sampling technique (SMOTE) with the random forest (RF) algorithm to address class imbalance and perform ensemble classification.
• Feature Composition: Uses four blocks of composed features derived from protein primary sequences to capture sequence characteristics influencing hydroxylation.
• Performance Metrics: Achieves Matthews correlation coefficient (MCC) values of 0.770 for hydroxyproline and 0.857 for hydroxylysine, representing improvements of 8% and 19% respectively over existing predictors as validated by jack-knife cross-validation.
• External Validation: Performance was confirmed using external datasets against other published methods.

Scientific Applications:

• Molecular mechanism analysis: Enables exploration of the molecular bases of diseases associated with abnormal protein hydroxylation.
• Disease-associated site discovery: Applied to human inherited diseases, identifying that loss of hydroxylation sites is more likely to cause disease than gain and revealing 52 human inherited diseases highly associated with loss of hydroxylation.

Methodology:

Feature extraction from protein primary sequences into four composed feature blocks; class imbalance handled with SMOTE; classification using random forest (RF); performance assessed by jack-knife cross-validation and validated on external datasets.