chemprop

Why we included this project

Turning molecular structures into reliable property predictions is a recurring practical problem for chemists and computational drug-discovery teams, and Chemprop is one of the most widely used libraries that tackles it. It relies on message-passing neural networks that learn directly from the molecular graph and SMILES input, so you skip hand-engineering feature vectors. Trained on curated datasets, it can predict endpoints like solubility or toxicity, and the clean CLI, Python API, and example notebooks get a first model running fast. The 2.x release reworked the package into a more modular, extensible design, making it easier to add custom featurizers or layers on top of the defaults. For small-molecule screening and ADMET profiling, it is a practical starting point that saves you from implementing the message-passing machinery yourself.