A deep learning toolkit for computational chemistry and drug design research with PyTorch backend.
OpenChem is a deep learning toolkit built with PyTorch for computational chemistry and drug design research. It provides modular components and a unified API to help researchers easily construct and train models for tasks like molecular property prediction, classification, regression, and generative modeling. The toolkit handles chemical data types like SMILES strings and molecular graphs, automating preprocessing and enabling faster experimentation.
Computational chemistry researchers, drug discovery scientists, and machine learning practitioners working on molecular modeling and chemical informatics problems.
OpenChem simplifies deep learning for chemistry by offering a configuration-based approach, reducing coding overhead while providing GPU-accelerated training and support for diverse chemical data types. Its modular design allows flexible model building, making it a practical alternative to building custom pipelines from scratch.
OpenChem: Deep Learning toolkit for Computational Chemistry and Drug Design Research
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Enables flexible model construction by easily combining modules like token embeddings, RNN encoders, and graph CNNs via a unified interface, as described in the main features.
Allows building new models using only configuration files, reducing coding overhead and speeding up experimentation, emphasized in the README's key features.
Supports multi-GPU training for fast iteration, requiring a modern NVIDIA GPU with CUDA 9.0+, as outlined in the installation requirements for performance.
Includes utilities for handling SMILES strings and automatic conversion to molecular graphs, detailed in the supported data types section for streamlined data handling.
Mandates a modern NVIDIA GPU with compute capability 3.5+ and CUDA 9.0, which can exclude researchers with limited or non-NVIDIA hardware resources.
Installation involves multiple steps including conda environments, RDKit dependencies, and potential Docker setup, making it less accessible for quick deployment.
The toolkit is still being populated with more models, so it may lack some advanced or specialized deep learning components for chemistry, as noted in the documentation.