BespokeFit and its dependencies can be installed in several ways. The OpenFF Initiative recommends using the
mamba package manager.
The recommended way to install
openff-bespokefit is via the
manager. A working installation also requires at least one package from each of
the two sections below (“Fragmentation Backends” and “Reference Data
mamba install -c conda-forge openff-bespokefit
If you do not have Mamba installed, see the OpenFF installation documentation.
AmberTools is free and open-source, and can generally be used fragment molecules up to 40 heavy atoms in under 10 minutes.
mamba install -c conda-forge ambertools
If you have access to the OpenEye toolkits (namely
oeomega) we recommend installing
these also as they can speed up certain operations significantly. OpenEye software requires a
free-for-academics license to run.
mamba install -c openeye openeye-toolkits
Reference Data Generators
Psi4 is an open source quantum chemistry package that enables BespokeFit to generate bespoke QC data, and is recommended to be installed unless you intend to train against data generated using a surrogate such as ANI:
mamba install -c psi4 -c conda-forge -c defaults psi4
There is an incompatibility between the AmberTools and Psi4 conda packages on Mac, and it is not possible to create a working conda environment containing both.
Installing Psi4 into an existing environment sometimes fails because of subtle differences in compiled dependencies found in multiple channels. An alternative is to install everything when initially creating the environment using, with AmberTools:
mamba create -n bespokefit-env -c psi4 -c conda-forge -c defaults python=3.9 openff-bespokefit psi4 ambertools
or with OpenEye Toolkits:
mamba create -n bespokefit-env -c psi4 -c conda-forge -c defaults -c openeye python=3.9 openff-bespokefit psi4 openeye-toolkits
xtb-python package gives access to the XTB semi-empirical models produced by the Grimme group, which may be
used as a much faster surrogate when generating QC reference data:
mamba install -c conda-forge xtb-python
xtb-python can optionally be configured to use MKL as its compute backend by running
mamba install -c conda-forge xtb-python "libblas=*=*mkl"
This likely provides better performance on Intel CPUs. Note that use of the MKL backend may be subject to additional license agreements with Intel. We currently understand it to be free for use by academics and companies generally, but it is not strictly open source.
TorchANI is a PyTorch implementation of the ANI neural network potentials from the Roitberg group that can be used as a much faster surrogate when generating QC reference data:
mamba install -c conda-forge torchani
TorchANI potentials are only suitable for molecules with a net neutral charge and have limited element coverage consisting of C, H, N, O, S, F and Cl
openff-bespokefit from source, begin by cloning the repository from
git clone https://github.com/openforcefield/openff-bespokefit cd openff-bespokefit
Create a custom conda environment which contains the required dependencies and activate it:
mamba env create --name openff-bespokefit --file devtools/conda-envs/test-env.yaml mamba activate openff-bespokefit
Finally, install the package itself into the new environment:
python setup.py develop