Installation

Installation#

Chemprop can either be installed from PyPI via pip or using a provided x.y.x_requirements.txt file, from source (i.e., directly from the git repo) using pip or the environment.yml file, or from Docker. The PyPI version includes the vast majority of Chemprop functionality, but some functionality is only accessible when installed from source. We recommend installing chemprop in a virtual environment (e.g., conda or miniconda). The following sections assume you are using conda or miniconda, but you can use any virtual environment manager you like (e.g. mamba).

Note

Python 3.11 vs. 3.12: Options 1, 2, and 4 below explicitly specify python=3.11 but you can choose to replace python=3.11 with python=3.12 in these commands. We test Chemprop on both versions in our CI.

Note

CPU-only installation: For the following options 1-3, if you do not have a GPU, you might need to manually install a CPU-only version of PyTorch. This should be handled automatically, but if you find that it is not, you should run the following command before installing Chemprop:

conda install pytorch cpuonly -c pytorch

Option 1: Installing from PyPI or Conda#

To install the latest version of Chemprop and all of its dependencies, execute the following commands:

conda create -n chemprop python=3.11
conda activate chemprop
pip install chemprop
# or
# conda install chemprop

Note

cuik-molmaker is a package that accelerates featurization of molecules using RDKit, and thereby accelerates training and inference. It can be installed via conda with conda install conda-forge::cuik_molmaker>=0.2. cuik-molmaker is automatically installed with Option 4: Installing via Docker. If installing with pip, use the optional dependency cuik_molmaker and the --extra-index-url https://pypi.nvidia.com/rdkit-latest/ flag, i.e., pip install chemprop[cuik_molmaker] --extra-index-url https://pypi.nvidia.com/rdkit-latest/. Additional RDKit versions are also supported, i.e., pip install chemprop[cuik_molmaker] --extra-index-url https://pypi.nvidia.com/rdkit-2025.09.1/. As of February 1, 2026 supported RDKit versions include 2025.03.5, 2025.3.6, and 2025.09.1. Support for additional versions may be checked by navigating to https://pypi.nvidia.com/ and searching for the RDKit version number.

The above approach is recommended as it will install the most updated versions of all external dependencies and will be more compatible with other dependencies you may wish to add. However, it is possible that external dependencies to Chemprop may introduce backwards incompatible changes to their codebase. For this reason, we also provide known working sets of dependencies for each Chemprop version. To install a specific version of Chemprop (i.e. 2.1.0), download the corresponding x.y.z_requirements.txt file from the Chemprop GitHub repository and run the following commands:

conda create -n chemprop python=3.11
conda activate chemprop
pip install -r x.y.z_requirements.txt

Option 2: Installing from source using pip#

conda create -n chemprop python=3.11
conda activate chemprop
git clone https://github.com/chemprop/chemprop.git
cd chemprop
pip install -e .

Note

You can also use this option to install additional optional dependencies by replacing pip install -e . with pip install -e ".[hpopt,dev,docs,test,notebooks,cuik_molmaker]" --extra-index-url https://pypi.nvidia.com/rdkit-latest/.

Option 3: Installing from source using environment.yml#

git clone https://github.com/chemprop/chemprop.git
cd chemprop
conda env create -f environment.yml
conda activate chemprop
pip install -e .

Option 4: Installing via Docker#

Chemprop can also be installed with Docker, making it possible to isolate the Chemprop code and environment. To install and run Chemprop in a Docker container, first install Docker. You may then either pull and use official Chemprop images or build the image yourself.

Note

The Chemprop Dockerfile runs only on CPU and does not support GPU acceleration. Linux users with NVIDIA GPUs may install the nvidia-container-toolkit from NVIDIA and modify the installation instructions in the Dockerfile to install the version of torch which is compatible with your system’s GPUs and drivers. Adding the --gpus all argument to docker run will then allow Chemprop to run on GPU from within the container. You can see other options for exposing GPUs in the Docker documentation. Users on other systems should install Chemprop from PyPI or source.

Pull Official Images#

docker pull chemprop/chemprop:X.Y.Z
docker run -it chemprop/chemprop:X.Y.Z

Where X, Y, and Z should be replaced with the version of Chemprop you wish to pull. For example, to pull chemprop-2.0.0 run

docker pull chemprop/chemprop:2.0.0

Note

Not all versions of Chemprop are available as pre-built images. Visit the Docker Hub page for a list of those that are available.

Note

Nightly builds of Chemprop are available under the latest tag on Dockerhub and are intended for developer use and as feature previews, not production deployment.

Build Image Locally#

See the build instructions in the top of the Dockerfile.