HEMDAG

HEMDAG predicts Gene Ontology (GO) terms for proteins using modular hierarchical ensemble methods to improve multi-class, multi-label protein function annotation.

Key Features:

• Modular hierarchical ensemble: HEMDAG employs a family of modular hierarchical ensemble methods that build upon existing flat classifiers to incorporate GO hierarchical structure into predictions.
• True-Path-Rule (TPR) compliance: Produces TPR-safe predictions consistent with the True-Path-Rule by enforcing hierarchy constraints.
• Isotonic regression and TPR learning: Uses isotonic regression and TPR learning strategies to correct and calibrate classifier outputs under GO constraints.
• Scalability: Scales to the full Gene Ontology and manages extensive datasets across multiple organisms.
• Competitive performance: Improves predictions from flat classifiers and competes with state-of-the-art hierarchy-aware methods, with evaluations reported in CAFA benchmarks.

Scientific Applications:

• Protein function annotation: Enhances automated assignment of GO terms in multi-class, multi-label protein function prediction tasks.
• Functional genomics: Supports large-scale annotation efforts that require GO-consistent functional labels.
• Proteomics: Improves functional interpretation of proteomic datasets by providing hierarchy-aware GO annotations.
• Systems biology: Provides GO-compliant functional labels useful for pathway and network analyses.

Methodology:

Constructs hierarchical ensembles of classifiers that enhance flat methods by incorporating GO hierarchical information and applies isotonic regression and TPR learning to ensure predictions adhere to the True-Path-Rule.