chemprop
Message Passing Neural Networks for Molecule Property Prediction
What is chemprop?
Message Passing Neural Networks for Molecule Property Prediction
Overview of chemprop
What it is
Chemprop is a repository that provides message passing neural networks for molecular property prediction. It is an open-source software package designed for machine learning on chemical data. The project underwent a ground-up rewrite and a major release (v2.0.0), with a transition guide available for users moving from v1 to v2. The repository includes documentation, tutorial notebooks in the examples/ directory, and references to the underlying theory and applications.
Key features
- Implements message passing neural networks for predicting molecular properties.
- Provides a command-line interface (CLI) for training and using models.
- Includes interpretability functionality, available as an example notebook in v2.
- Supports atom and bond-level predictions.
- Offers pre-trained model checkpoints for selected applications (e.g., antibiotic discovery).
- Provides a transition guide from v1 to v2, including side-by-side CLI argument comparisons and changes to default hyperparameters.
Use cases
Chemprop has been used in several research applications. It was used to predict antibiotic activity against E. coli, leading to the discovery of Halicin, a novel antibiotic candidate. It was also used to identify a structural class of antibiotics selective against methicillin-resistant Staphylococcus aureus (MRSA) and vancomycin-resistant enterococci, with results explained using Chemprop's interpret method. Additionally, Chemprop was trained on 41 ADMET (absorption, distribution, metabolism, excretion, and toxicity) datasets from the Therapeutics Data Commons to create the ADMET-AI platform.
Getting started
See the repository README for installation and quick-start instructions. The README notes that tutorial notebooks are available in the examples/ directory.
License and pricing
Chemprop is free to use under the MIT License. The Chemprop logo is free to use under CC0 1.0. The software is open source and no pricing information is stated.
Based on the project's public README and repository metadata.