GasPhos

GasPhos predicts kinase-specific human protein phosphorylation sites to support analysis of phosphorylation-mediated processes such as DNA repair, transcriptional regulation, and signal transduction.

Key Features:

• Kinase-specific prediction: Predicts human phosphorylation sites across six kinases and reports superior predictive performance compared to existing phosphorylation prediction tools.
• Gas feature selection: Implements the "Gas" hybrid feature selection approach that integrates a genetic algorithm (GA) with an ant colony system.
• MDGI heuristic: Uses the mean decrease Gini index (MDGI) as a heuristic measure for path selection within the ant colony framework.
• Binary transformation and state transitions: Incorporates binary transformation strategies and novel state transition rules to refine feature selection.
• Performance evaluation and model selection: Employs performance evaluation strategies focused on individual kinases to select optimal learning models tailored to each kinase.
• Disease-related phosphorylation analysis: Analyzes disease-related phosphorylated proteins to provide insights into pathological mechanisms and potential therapeutic targets.
• Method versatility: Applies the Gas feature selection approach beyond phosphorylation prediction to other feature selection problems.

Scientific Applications:

• Phosphorylation mapping: Mapping kinase-specific phosphorylation sites for studies of DNA repair, transcriptional regulation, and signal transduction.
• Disease association: Prioritizing phosphorylated proteins implicated in disease and identifying potential therapeutic targets.
• Feature selection in bioinformatics: Applying the Gas feature selection method to other biological classification and feature selection tasks.

Methodology:

Hybrid feature selection combining a genetic algorithm (GA) with an ant colony system; mean decrease Gini index (MDGI) used for ant path selection; incorporation of binary transformation strategies and novel state transition rules; performance evaluation per kinase for optimal learning-model selection.