STALLION

STALLION predicts prokaryotic protein lysine acetylation (Kace) sites using a stacking-based ensemble learning approach to improve site identification across six prokaryotic species.

Key Features:

• Target biological problem: Predicts lysine acetylation (Kace) sites in prokaryotic proteins across six different species.
• Stacking-based ensemble: Integrates multiple machine learning models via a stacking strategy to refine prediction accuracy.
• Species-specific models: Constructs and optimizes models independently for each of the six prokaryotic species.
• Feature encodings: Employs 11 distinct encodings that capture three key characteristics surrounding Kace sites.
• Feature selection: Applies a systematic, species-specific feature selection strategy to determine optimal feature sets.
• Baseline algorithms: Builds baseline models using five different tree-based ensemble algorithms.
• Meta-classifier: Combines baseline model outputs and trains a higher-level classifier as the stacking meta-learner.
• Benchmarking: Demonstrates superior performance compared to existing predictors on independent test datasets.

Scientific Applications:

• Kace site prediction: Identification of candidate lysine acetylation sites in prokaryotic proteomes.
• Cross-species analysis: Comparative modeling of acetylation patterns across six prokaryotic species.
• Experimental prioritization: Prioritization of candidate sites for experimental validation in studies of prokaryotic protein modifications.
• Method benchmarking: Comparative evaluation and benchmarking of computational predictors on independent test datasets.

Methodology:

Uses 11 sequence encodings representing three local characteristics, applies systematic species-specific feature selection, trains five tree-based ensemble baseline models per species, and combines their outputs with a stacking meta-classifier.