Sirtuin Functional Classification

Instead of using a single multiclass classifier, we introduce a novel ensemble strategy: four independently trained binary models, each specialized to distinguish one functional Sirtuin class against all others.

This allows each model to learn class-specific biological signals, increasing sensitivity and interpretability per class.

During prediction, each file is evaluated by all four models. Their outputs are compared against optimal thresholds, and a voting mechanism selects the best match based on confidence and margin.

This strategy consistently outperformed multiclass models in accuracy, MCC, and AUC across all feature types.