UbPred

UbPred predicts ubiquitination sites on lysine residues in proteins to identify probable post-translational ubiquitin modifications relevant to protein regulation and disease.

Key Features:

• Predictive Accuracy: UbPred achieves a class-balanced accuracy of 72% and an area under the ROC curve (AUC) of 80% for ubiquitination site prediction.
• Machine Learning Model: The predictor is based on a random forest-based predictive model trained on experimentally derived ubiquitination sites.
• Training Data: The training dataset comprises newly identified and previously known ubiquitination sites obtained using liquid chromatography, mass spectrometry, and mutant Saccharomyces cerevisiae strains.
• Sequence and Structural Biases: Analyses of sequence biases and structural preferences around sites indicate characteristics resembling intrinsically disordered protein regions.
• Proteome-wide Predictions: Proteome-wide application reveals enrichment of predicted ubiquitination in transcription/enzyme regulators and cell cycle control proteins in Saccharomyces cerevisiae and in cytoskeletal, cell cycle, regulatory, and cancer-associated proteins in the human proteome.
• Ligase and Half-life Enrichment: High-confidence predicted sites are enriched among substrates of the Rsp5 ubiquitin ligase and in proteins with very short half-lives.
• Disease Association: Changes in predicted ubiquitination sites are proposed as potential molecular mechanisms underlying disease-associated mutations.

Scientific Applications:

• Proteome-wide mapping: Generate proteome-scale maps of candidate lysine ubiquitination sites in Saccharomyces cerevisiae and human proteomes.
• Prioritizing ligase substrates: Identify and prioritize candidate substrates for ubiquitin ligases such as Rsp5.
• Linking ubiquitination to protein stability: Investigate associations between predicted ubiquitination sites and protein half-life or regulatory turnover.
• Interpreting disease variants: Assess the potential impact of mutations on predicted ubiquitination sites to inform studies of disease-associated variants.

Methodology:

UbPred uses a random forest-based predictive model trained on a dataset of newly identified and known ubiquitination sites generated using liquid chromatography and mass spectrometry from mutant Saccharomyces cerevisiae strains, with analyses of sequence biases and structural preferences indicating similarity to intrinsically disordered regions.