A convolutional neural network and graph convolutional network-based method for predicting the classification of...

Area of Science:

• Pharmacology
• Computational Chemistry
• Bioinformatics

Background:

• The World Health Organization's Anatomical Therapeutic Chemical (ATC) system is crucial for classifying medicines.
• Accurate ATC classification aids in identifying compound properties and potential therapeutic applications in drug discovery.

Purpose of the Study:

• To develop an advanced computational model for predicting the 14 main ATC classes for chemical compounds.
• To enhance the accuracy and efficiency of drug classification in the pharmaceutical research landscape.

Main Methods:

• Developed CGATCPred, an end-to-end multi-label classifier integrating Convolutional Neural Networks (CNNs) and Graph Convolutional Networks (GCNs).
• Utilized deep CNNs with shortcut connections for feature extraction and GCNs on ATC class correlation graphs for label embedding abstraction.
• Employed label embeddings to guide the compound representation learning process.

Main Results:

• CGATCPred achieved high performance metrics, including Aiming of 81.94%, Coverage of 82.88%, and Accuracy of 80.81% in predicting ATC classes.
• Demonstrated significant improvements over existing multi-label classification methods.
• The model's effectiveness was validated using a Jackknife test.

Conclusions:

• CGATCPred offers a robust and accurate computational approach for multi-label classification of chemical compounds according to the ATC system.
• The developed method provides valuable tools for drug discovery by inferring therapeutic and pharmacological properties.
• The model's code is publicly available, facilitating further research and application in medicinal chemistry.