Predicting combinative drug pairs towards realistic screening via integrating heterogeneous features

Area of Science:

• Computational drug discovery
• Pharmacology
• Bioinformatics

Background:

• Drug combinations are crucial for treating complex diseases but identifying them is costly.
• Current computational methods struggle with integrating diverse data and predicting combinations involving new drugs.

Purpose of the Study:

• To develop a novel drug-driven approach for large-scale prediction of potential drug combinations.
• To address limitations in feature integration and prediction for new drugs.

Main Methods:

• Defined four novel features using heterogeneous data.
• Developed an efficient feature fusion scheme.
• Implemented appropriate cross-validation for realistic screening scenarios.

Main Results:

• The fusion scheme effectively integrates heterogeneous features.
• The approach demonstrates strong predictive power in three screening scenarios: known drugs (AUC 0.954, AUPR 0.821), known and new drugs (AUC 0.909, AUPR 0.635), and new drugs (AUC 0.809, AUPR 0.592).

Conclusions:

• The developed approach offers an effective tool for integrating heterogeneous features in drug combination prediction.
• It is the first tool capable of predicting potential combinative pairs among new drugs.