An enhanced cascade-based deep forest model for drug combination prediction.

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Drug combination therapy offers significant potential for complex diseases.
• The vast search space of drug combinations hinders experimental validation.
• Machine learning methods are increasingly used to discover synergistic drug combinations and reduce experimental workload.

Purpose of the Study:

• To predict novel synergistic drug combinations in various cancer cell lines.
• To incorporate cell line-specific drug-induced gene expression profiles (GP) as a feature type.
• To reveal the biological mechanisms underlying synergistic drug effects.

Main Methods:

• An enhanced cascade-based deep forest regressor (EC-DFR) was developed.
• The EC-DFR model utilizes a small-scale drug combination dataset including chemical, physical, and biological (GP) properties.
• The model's performance was compared against state-of-the-art deep neural networks and classical machine learning algorithms.

Main Results:

• EC-DFR demonstrated superior performance compared to existing methods in predicting synergistic drug combinations.
• Biological experimental validation confirmed the model's predictive accuracy on novel drug combinations.
• Gene expression profiles (GP) were identified as the most crucial feature type, contributing 82.40% to the prediction accuracy.

Conclusions:

• The EC-DFR model effectively predicts synergistic drug combinations by integrating gene expression profiles.
• Cellular responses captured by gene expression profiles are critical for predicting drug synergism.
• The model's interpretability allows for the identification of key genes and potential relationships influencing drug combination synergy.