DGL-LifeSci

DGL-LifeSci provides utilities to apply graph neural networks to chemical and biological graph-structured data for molecular property prediction, reaction prediction, and drug–target interaction prediction.

Key Features:

• Graph Neural Network framework: Modules for representing and modeling molecular graphs to support molecular property prediction, reaction prediction, and drug–target interaction prediction.
• Integration with RDKit, PyTorch, and DGL: Uses RDKit for cheminformatics and implements GNN models on the PyTorch and DGL frameworks.
• Performance optimization and benchmarks: Evaluated on MoleculeNet, USPTO, and ZINC and reported up to six times faster processing compared to previous implementations.
• Modeling flexibility and pretrained models: Provides optimized modules for different modeling stages and pretrained models for reproducing benchmark experiments or applying models without retraining.
• Data preprocessing modules: Includes preprocessing routines tailored for molecular graph data and graph-structured workflows.

Scientific Applications:

• Molecular Property Prediction: Predicts physicochemical and biological properties of molecules to support drug discovery and materials research.
• Reaction Prediction: Predicts chemical reaction outcomes to assist in compound design and reaction mechanism studies.
• Drug–Target Interaction Prediction: Predicts interactions between small molecules and biological targets to aid therapeutic discovery.

Methodology:

Data preprocessing tailored for molecular graphs; implementation of graph neural network models using PyTorch and DGL; utilization of RDKit for cheminformatics tasks.