---
id: chemprop
version: "2.3.1"
license: MIT
license_treatment: permissive
maintenance: active
---

# chemprop — Molecular Property Prediction with Message Passing Neural Networks

License: permissive · Maintenance: active · Downloads: 110.9K/mo

## What it is and what it does

Chemprop is a machine learning framework for predicting molecular properties using graph neural networks. It represents molecules as graphs and applies message passing to learn structural patterns that correlate with chemical properties, enabling applications in drug discovery, ADMET prediction, and molecular optimization.

The package underwent a major rewrite in v2.0.0 and now provides a modular, PyTorch-Lightning-based architecture. It depends on torch, lightning, numpy, pandas, rdkit, scikit-learn, and scipy, plus specialized chemistry libraries like descriptastorus and astartes. The framework is actively maintained, supports Python 3.11–3.14, and is permissively licensed under MIT.

Use it for:

- Train models to predict absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties for drug candidates
- Discover novel antibiotics by predicting bioactivity against bacterial strains from molecular structures
- Build ensemble models for structure–activity relationship (SAR) analysis with interpretability
- Predict reaction properties or atom/bond-level predictions for chemical synthesis planning
- Benchmark molecular representations and compare graph neural network architectures on property datasets

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Chemprop trains and deploys message passing neural networks to predict molecular properties from chemical structures, using graph neural networks to learn from molecular representations.

Yes, if you work in computational chemistry, drug discovery, or molecular ML. Chemprop is actively maintained, well-cited in peer-reviewed research, and has a permissive license. The dependency stack is substantial but standard for this domain. Install friction is low. No known vulnerabilities. The v2 rewrite is recent; verify that your use case is covered in current documentation.

## Install

pip install chemprop
uv add chemprop
poetry add chemprop

## Installing chemprop

Before you install: Low install friction with a pure-Python wheel. Active maintenance with a release 10 days ago and commits through 2026-08-04. Requires 13 runtime dependencies including torch, lightning, rdkit, and scikit-learn—a substantial but standard stack for molecular ML work.

License in practice: MIT license permits commercial and private use with minimal restrictions, making it suitable for research, industry applications, and derivative work.

Quickstart:

pip install chemprop

import chemprop
from chemprop import Chemprop

# Train a model on molecular SMILES and property data
model = Chemprop()
model.train(smiles_list, property_values)

Requires PyTorch and RDKit; GPU support optional but recommended for large datasets. Python 3.11–3.14 only.

Verify before relying:

- Exact API surface and whether the snippet example reflects current v2.3.1 usage patterns
- Whether all 13 runtime dependencies are required for basic property prediction or only for advanced features
- Performance characteristics and typical training time for datasets of different sizes

## Package facts

- License: MIT (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 110.9K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

molecular property prediction, graph neural networks chemistry, drug discovery machine learning, chemical structure prediction, message passing neural networks, MPNN molecular modeling, cheminformatics deep learning, molecular-ml, drug-discovery, graph-neural-networks

[View on SkillFed](https://skillfed.io/packages/chemprop) · [View on PyPI](https://pypi.org/project/chemprop/)