Machine learning model for anti-cancer drug combinations: Analysis, prediction, and validation

Area of Science:

• Oncology
• Pharmacology
• Computational Biology

Background:

• Drug combination therapy enhances anti-cancer efficacy and overcomes resistance.
• Screening all possible drug combinations is impractical due to vast numbers.
• Limited biological understanding of synergistic drug pairs hinders rational selection.

Purpose of the Study:

• To systematically analyze drug combination data to identify synergistic pairs.
• To develop and validate machine learning models for predicting synergistic drug combinations.
• To uncover potential biomarkers for synergistic drug pairs.

Main Methods:

• Systematic analysis of drug combination datasets from multiple databases.
• Classification of drug pairs based on Mechanism of Action (MoA).
• Development and validation of machine learning models using 2D cell lines and 3D tumor slice cultures (3D-TSC).
• Incorporation of molecular features and protein-protein interaction networks for biomarker identification.

Main Results:

• Identified 110 MoA pairs significantly enriched in synergy across various cancers.
• Machine learning models achieved improved predictive performance for synergistic drug effects.
• Validated the combination of Lapatinib and Pazopanib as therapeutically effective in breast cancer by targeting the PI3K/AKT/mTOR pathway.
• Identified potential biomarkers linked to drug targets via protein-protein interactions.

Conclusions:

• This study provides a data-driven approach to predict synergistic drug combinations for cancer therapy.
• Machine learning models offer practical utility for identifying effective drug pairs, validated through experimental models.
• The findings guide rational drug combination development and highlight Lapatinib-Pazopanib as a promising therapeutic strategy for breast cancer.