LinkPhinder

LinkPhinder predicts kinase–substrate relationships by applying statistical relational learning to phosphorylation networks represented as knowledge graphs to enable large-scale inference of phosphorylation events and comprehensive kinome coverage.

Key Features:

• Knowledge-graph representation: Phosphorylation data are encoded as knowledge graphs to represent networked biological relationships.
• Statistical relational learning: Uses statistical relational learning algorithms to infer kinase–substrate interactions from the knowledge graph.
• Network-based AI approach: Leverages a network-based artificial intelligence framework rather than local-feature-only models.
• Large-scale kinome coverage: Enables broad coverage across the kinome for comprehensive prediction of phosphorylation events.
• High-confidence predictions: Produces high-confidence predicted kinase–substrate relationships suitable for experimental follow-up.
• Benchmarking: Demonstrated superior kinome coverage relative to six existing prediction systems.
• Experimental validation: Predictions have been experimentally validated for phosphorylations involving human kinases LATS1, AKT1, PKA, and MST2.

Scientific Applications:

• Guiding phosphoproteomics: Prioritizes candidate phosphorylation sites and kinase assignments for phosphoproteomic experiments.
• Discovery of novel phosphorylation reactions: Facilitates identification of previously unknown kinase–substrate relationships.
• Mapping signaling networks: Supports reconstruction and analysis of protein signaling networks involved in cell-fate decisions and other cellular processes.
• Translational research: Provides predictions that can inform studies in biology, medicine, and drug development focused on kinase signaling.

Methodology:

Applies statistical relational learning to phosphorylation networks encoded as knowledge graphs to predict kinase–substrate interactions.