Note: this site is several versions behind. An up-to-date version of our Read the Docs is forthcoming with the release of Chemprop v2.0 in early 2024. The README and args.py files are currently the best sources for documentation on more recently-added features.
Chemprop is a message passing neural network for molecular property prediction.
At its core, Chemprop contains a directed message passing neural network (D-MPNN), which was first presented in Analyzing Learned Molecular Representations for Property Prediction. The Chemprop D-MPNN shows strong molecular property prediction capabilities across a range of properties, from quantum mechanical energy to human toxicity.
Chemprop was later used in the paper A Deep Learning Approach to Antibiotic Discovery to discover promising new antibiotics by predicting the likelihood that a molecule would inhibit the growth of E. coli.
- Web Interface
- Training and Predicting
- Hyperparameter Optimization
- Command Line Arguments
- Neural Network Utility Functions
- Utility Functions
- Scikit-Learn Models
- Useful Scripts