A dual granular balanced deep forest model for effective drug combination prediction

Area of Science:

• Computational Biology
• Pharmacology
• Machine Learning

Background:

• Combination therapies are crucial for treating complex diseases.
• Identifying synergistic drug pairs is challenging due to vast search spaces and imbalanced data.

Purpose of the Study:

• To develop a deep-forest framework for improved drug synergy prediction.
• To address challenges posed by highly skewed class distributions in synergy data.

Main Methods:

• Utilized a deep-forest model prioritizing informative and uncertain training examples.
• Compared the framework against canonical, imbalanced learning, and drug-specific models.
• Performed model interpretability analyses to identify key chemical and genetic features.

Main Results:

• The proposed method consistently outperformed existing prediction models.
• Achieved favorable results despite highly skewed class distributions.
• Identified specific chemical substructures and cell-line genetic signals associated with drug synergy.

Conclusions:

• The deep-forest framework offers a practical and interpretable approach to synergy prediction.
• The method accelerates the discovery of biologically plausible synergistic drug combinations.
• Enhances the identification of effective combination therapies for complex diseases.