KinasePhos 3.0

KinasePhos 3.0 predicts kinase-specific phosphorylation sites to support identification of phosphorylation-mediated regulation and signaling relationships in proteins.

Key Features:

• Experimental dataset: Aggregates 41,421 experimentally verified kinase-specific phosphorylation sites.
• Data sources: Integrates phosphorylation data from PhosphoSitePlus, UniProt, GPS 5.0, and Phospho.ELM.
• Kinase coverage: Identifies 1,380 unique kinases with 753 kinases classified in KinBase and 627 annotated via a phylogenetic tree.
• Predictive models: Provides 771 predictive models at individual, family, and group levels trained on datasets containing at least 15 verified substrate sites each.
• Machine learning: Trains models using machine learning methodologies.
• Interpretability: Applies Shapley additive explanations (SHAP) to assess feature importance.
• Predictive performance: Reports accuracies of 94.5% for Akt family sites, 92.5% for CKT family sites, 90.0% for PKA family sites, and an average accuracy of 87.2% across models.

Scientific Applications:

• Signaling pathway analysis: Enables identification and mapping of kinase–substrate relationships to elucidate signaling pathways.
• Therapeutic target identification: Supports discovery of potential kinase-related therapeutic targets through predicted phosphorylation sites.
• Protein regulation and proteomics: Facilitates investigation of protein function regulation via phosphorylation in molecular biology and proteomics studies.

Methodology:

Aggregated 41,421 experimentally verified kinase-specific phosphorylation sites from PhosphoSitePlus, UniProt, GPS 5.0, and Phospho.ELM; identified 1,380 unique kinases (753 from KinBase and 627 annotated via phylogenetic tree); constructed 771 predictive models at individual, family, and group levels trained on datasets with at least 15 verified substrate sites; used machine learning methodologies and Shapley additive explanations (SHAP) for feature importance.