PhosphoPredict

PhosphoPredict predicts kinase-specific phosphorylation substrates and sites within the human proteome using protein sequence and functional features.

Key Features:

• Kinase-Specific Predictions: Predicts substrates and phosphorylation sites for specific kinases to annotate kinase–substrate relationships within the human proteome.
• Kinase Coverage: Targets 12 human kinases and kinase families, including ATM, CDKs, GSK-3, MAPKs, PKA, PKB, PKC, and SRC.
• Feature Integration: Integrates protein sequence and functional features and identifies critical determinants most informative for substrate specificity across kinase families.
• Machine Learning Model: Constructs prediction models using random forest (RF) algorithms.
• Benchmarking and Validation: Evaluated by five-fold cross-validation and independent tests and shown to perform comparably to KinasePhos, PPSP, GPS, and Musite.
• Improved Accuracy: Combination of sequence and functional features enhances phosphorylation site prediction accuracy across targeted kinases.

Scientific Applications:

• High-Throughput Identification: Enables high-throughput identification of kinase-specific phosphorylation sites for basic and translational research.
• Proteome-Wide Analysis: Applied to the entire human proteome to predict approximately 150 to 800 potential substrates per kinase or kinase family.

Methodology:

Random forest (RF) algorithms analyze integrated protein sequence and functional features with selection of informative feature subsets; models were evaluated by five-fold cross-validation and independent tests.