DeepNC

DeepNC predicts drug-target binding affinities and interactions by applying graph neural networks to molecular graph representations and 1-D convolutional networks for target features to improve drug-target interaction (DTI) prediction.

Key Features:

• Graph Neural Network Integration: Implements GENConv, GCNConv, and HypergraphConv to model molecular structures as graphs and capture chemical and structural nuances of compounds.
• Multi-Layered Representation Learning: Uses stacked GNN layers to learn drug features and 1-D convolutional networks to learn target features for joint representation learning.
• Predictive Modeling: Combines learned drug and target representations in fully-connected layers to predict binding affinity values for DTI assessment.

Scientific Applications:

• Drug-Target Interaction Prediction: Predicts binding affinities to support identification of compound-target pairs in computational biology and pharmacology.
• Drug Discovery Prioritization: Facilitates prioritization of candidate compounds for experimental validation by improving DTI prediction accuracy.

Methodology:

Molecules are represented as graphs; GNNs (GENConv, GCNConv, HypergraphConv) learn drug features while 1-D convolutional networks learn target features; learned representations are integrated via fully-connected layers to predict binding affinity and validated on Davis, Kiba, and Allergy datasets using mean square error and concordance index.