CGATCPred

CGATCPred predicts Anatomical Therapeutic Chemical (ATC) classifications of chemical compounds using deep learning models that integrate convolutional neural networks (CNNs) and graph convolutional networks (GCNs).

Key Features:

• Multi-Label ATC Classification: Predicts compound membership across multiple Anatomical Therapeutic Chemical (ATC) classes using a multi-label classification framework.
• CNN-Based Compound Feature Extraction: Utilizes deep convolutional neural networks with shortcut connections to extract compound features from seven association scores between compounds.
• ATC Class Correlation Modeling: Constructs a correlation graph of ATC classes and applies graph convolutional networks to generate label embeddings capturing relationships among classes.

Scientific Applications:

• Drug Classification and Annotation: Assigns compounds to Anatomical Therapeutic Chemical (ATC) classes to infer therapeutic, pharmacological, and chemical properties.
• Computational Drug Discovery: Supports identification of potential active compounds and prediction of therapeutic categories during drug development.

Methodology:

CGATCPred applies an end-to-end multi-label classification framework that combines convolutional neural networks for compound feature extraction with graph convolutional networks operating on an ATC class correlation graph to learn label embeddings and guide compound classification.