DGL-LifeSci: An Open-Source Toolkit for Deep Learning on Graphs in Life Science.

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning

Background:

• Graph neural networks (GNNs) are powerful for analyzing molecular data in chemistry and biology.
• GNN-based modeling presents challenges in data preprocessing, programming, and deep learning expertise.
• Existing tools often require significant technical background, limiting accessibility.

Purpose of the Study:

• Introduce Deep Graph Library (DGL)-LifeSci, an open-source Python toolkit for GNNs in life sciences.
• Enable GNN-based modeling for molecular property prediction, reaction prediction, and molecule generation.
• Provide user-friendly command-line interfaces for researchers without programming or deep learning backgrounds.

Main Methods:

• Developed DGL-LifeSci as a Python toolkit integrating RDKit, PyTorch, and Deep Graph Library (DGL).
• Implemented command-line interfaces for streamlined GNN model execution on custom datasets.
• Validated performance on standard benchmarks: MoleculeNet, USPTO, and ZINC.

Main Results:

• DGL-LifeSci achieves up to a 6x speedup compared to previous implementations.
• The toolkit offers optimized modules for flexibility across the modeling pipeline.
• Pretrained models are available for result reproduction and immediate application.

Conclusions:

• DGL-LifeSci democratizes GNN application in life science research by simplifying complex modeling tasks.
• The package enhances efficiency and accessibility for molecular property and reaction prediction.
• Open-source availability and pretrained models facilitate broader adoption and reproducibility.