UbiSite

UbiSite predicts ubiquitin-conjugation sites on lysine residues in proteins to support analysis of protein ubiquitylation mediated by E1, E2, and E3 enzymes.

Key Features:

• Two-Layered Machine Learning Approach: UbiSite employs a two-layered support vector machine (SVM) model that integrates substrate motifs to improve prediction accuracy of ubiquitin-conjugation sites.
• Motif Discovery and Integration: Substrate motifs are identified with MDDLogo from ubiquitylation datasets and incorporated into the prediction model.
• Feature Integration: The model uses amino acid composition (AAC), positional weighted matrix (PWM), position-specific scoring matrix (PSSM), and solvent-accessible surface area (SASA) as input features.
• PSSM Discriminative Power: Position-specific scoring matrix (PSSM) was identified as the most effective feature for distinguishing ubiquitylation from non-ubiquitylation sites.
• Cross-Validation Performance: Five-fold cross-validation of the two-layered SVM integrating MDDLogo motifs achieved 81.06% accuracy and a Matthews Correlation Coefficient (MCC) of 0.586.
• Independent Test Performance: Independent testing reported sensitivity 85.10%, specificity 69.69%, accuracy 73.69%, and MCC 0.483.

Scientific Applications:

• Large-scale ubiquitinome studies: Identification of lysine ubiquitylation sites across proteomes using motif-aware predictions.
• Analysis of ubiquitylation-mediated regulation: Support for investigating the regulatory roles of protein ubiquitylation in cellular functions.

Methodology:

Two-layered SVM model integrating MDDLogo-identified substrate motifs using AAC, PWM, PSSM, and SASA features, evaluated by five-fold cross-validation and independent testing.