MOCT: A Multi-Class Oblique Tree Algorithm for Synergistic Drug Combination Prediction.

Area of Science:

• Computational biology
• Machine learning
• Pharmacology

Background:

• Machine learning models are increasingly used for drug combination prediction.
• Class imbalance and lack of interpretability are significant challenges in current predictive models.
• Existing methods struggle to address these issues effectively in complex biological datasets.

Purpose of the Study:

• To propose a novel clustering-based oblique decision tree (MOCT) algorithm.
• To extract interpretable knowledge from multi-class datasets, specifically for drug combination prediction.
• To address the limitations of traditional methods in handling class imbalance and ensuring model interpretability.

Main Methods:

• Developed a clustering-based oblique decision tree (MOCT) algorithm.
• Implemented a hierarchical clustering approach to group samples by class.
• Generated feature subspaces for data splitting and node creation, optimizing for interpretability and conciseness.
• MOCT grows only one non-leaf node per layer, ensuring a compact tree structure.

Main Results:

• The MOCT algorithm demonstrated superior performance compared to other methods in drug combination prediction tasks.
• The proposed method effectively handled class imbalance issues inherent in biological datasets.
• Experimental results from three cell lines showed enhanced model interpretability, crucial for biological and medical experts.

Conclusions:

• The MOCT algorithm offers a robust solution for drug combination prediction, particularly in scenarios with class imbalance.
• The interpretability of MOCT facilitates better understanding and trust in predictive models for clinical applications.
• This approach advances the application of machine learning in precision medicine by providing interpretable and accurate predictions.