A model with deep analysis on a large drug network for drug classification.

Area of Science:

• Pharmacology and Cheminformatics
• Computational Drug Discovery
• Network Science in Biology

Background:

• Drug classification is crucial for understanding therapeutic effects and guiding novel drug discovery.
• The Kyoto Encyclopedia of Genes and Genomes (KEGG) DRUG system categorizes drugs into 14 classes based on shared properties.
• Accurate drug classification aids in predicting potential therapeutic applications and accelerating research.

Purpose of the Study:

• To develop and evaluate a novel drug classification model.
• To assign drugs to one of the 14 classes within the KEGG DRUG classification system.
• To explore the utility of network-derived features for drug classification.

Main Methods:

• Drugs were represented using novel features derived from a large drug network via the Node2vec network embedding algorithm.
• The synthetic minority over-sampling technique (SMOTE) was applied to address class imbalance in the dataset.
• A support vector machine (SVM) classifier was trained on the balanced dataset.

Main Results:

• The proposed classification model demonstrated excellent performance, validated by 10-fold cross-validation.
• The model significantly outperformed methods utilizing traditional fingerprint features and other network embedding algorithms.
• The inclusion of SMOTE resulted in more balanced performance across all drug classes.

Conclusions:

• Network-derived features, particularly from Node2vec, are effective for drug classification.
• The developed model offers a robust and superior approach compared to existing methods.
• This classification system has the potential to accelerate the identification of novel drug candidates.