Predicting effective drug combinations using gradient tree boosting based on features extracted from drug-protein...

Area of Science:

• Computational biology
• Network pharmacology
• Drug discovery

Background:

• Targeted cancer therapies face limitations due to drug resistance.
• Drug combinations offer synergistic effects by targeting multiple pathways.

Purpose of the Study:

• To develop a computational method for predicting effective drug combinations.
• To enhance cancer treatment strategies by overcoming drug resistance.

Main Methods:

• Constructed a heterogeneous drug-protein network integrating drug/protein similarity and known associations.
• Applied Random Walk with Restart (RWR) to generate feature vectors for drug combinations.
• Trained a Gradient Tree Boosting (GTB) classifier to predict novel drug combinations.

Main Results:

• The proposed method achieved superior performance compared to seven other classifiers.
• Heterogeneous network-derived features proved more informative than traditional ontology features.
• The approach demonstrated effectiveness in predicting synergistic drug combinations.

Conclusions:

• The developed method offers a systematic and reliable approach for drug discovery.
• Network pharmacology principles are effectively utilized to exploit target interactions.
• This strategy holds promise for improving cancer patient outcomes.