Five-Feature Model for Developing the Classifier for Synergistic vs. Antagonistic Drug Combinations Built by XGBoost

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Combinatorial drug therapy offers improved efficacy and reduced adverse events.
• In silico methods for classifying drug interactions are more efficient than experimental approaches.
• Existing computational methods are predominantly applied to cancer therapies.

Purpose of the Study:

• To introduce a novel XGBoost-based computational method for classifying synergistic versus antagonistic drug combinations.
• To evaluate the performance of the XGBoost model using five drug and biomolecular network features.
• To assess the model's applicability across different drug categories beyond cancer.

Main Methods:

• Development of an XGBoost classifier utilizing five key features related to drugs and their target biomolecular networks.
• Stratified fivefold cross-validation and independent validation datasets were employed for model assessment.
• Comparison of XGBoost performance against logistic regression, naïve Bayesian, and random forest classifiers.
• Validation on The Cancer Genome Atlas (TCGA) data and analysis across WHO Anatomic Therapeutic Class groups.

Main Results:

• XGBoost demonstrated superior predictive accuracy compared to other models in both cross-validation and independent validation (e.g., 0.86 accuracy for XGBoost vs. 0.78 for others on an independent set).
• The five-feature XGBoost model showed higher effectiveness in predicting synergistic drug combinations than antagonistic ones.
• Validation on TCGA data yielded an accuracy of 0.79 for 61 drug pairs, comparable to existing methods like DeepSynergy.
• Significant changes in prediction accuracy were observed for drugs within specific anatomical/pharmacological groups (Fisher's exact test, p < 0.05).

Conclusions:

• The five-feature XGBoost model presents a robust and efficient computational strategy for classifying synergistic and antagonistic drug combinations.
• This method offers significant benefits for predicting drug interactions across diverse therapeutic areas.
• The model's performance suggests its potential utility in drug discovery and development pipelines.