Could not install

[PatWalters]

Your package looks very interesting; however, I was unable to install it on either Linux or macOS. Your environment.yml seems quite lengthy and overly specific. Does your package really require all of these dependencies? I would recommend starting from a bare-bones Python installation and building an environment with only the packages necessary to run your application. Then, use that environment to create your environment.yml.

[greglandrum]

Yeah, I was going to suggest this too.

@katzberger , I can show you how to do this

[corinwagen]

We're interested in trying to combine these implicit solvent models w/ DFT/NNPs and run some benchmarks, but are also struggling with incorporating the current repository.

Would it be possible to (1) expose a simple interface for just getting the solvent energy + forces from a given topology + coordinates and (2) put this package, or some form thereof, onto pypi or conda? This would help make it simpler to mix-and-match GNNIS with e.g. AIMNet2 or Psi4 or PySCF and an external optimizer.

We're happy to contribute to either of these efforts if we can be helpful.

[greglandrum]

We're interested in trying to combine these implicit solvent models w/ DFT/NNPs and run some benchmarks, but are also struggling with incorporating the current repository.

Hi @corinwagen we've already done some work on integrating GNNIS with QM that may be helpful to you.
The repo for that is here:
https://github.com/rinikerlab/QM-GNNIS
and the associated preprint (still in review) is:
https://doi.org/10.26434/chemrxiv-2025-sx29m

Would it be possible to (1) expose a simple interface for just getting the solvent energy + forces from a given topology + coordinates and (2) put this package, or some form thereof, onto pypi or conda? This would help make it simpler to mix-and-match GNNIS with e.g. AIMNet2 or Psi4 or PySCF and an external optimizer.

We're happy to contribute to either of these efforts if we can be helpful.

We will take a look at figuring out whether or not we can do a conda-forge package; in the meantime the best we can offer is installing this with pip from github:
python -m pip install git+https://github.com/rinikerlab/GNNImplicitSolvent
after you have setup an environment with the dependencies. We do still need to improve environment.yml (as discussed above)

[greglandrum]

Your package looks very interesting; however, I was unable to install it on either Linux or macOS. Your environment.yml seems quite lengthy and overly specific. Does your package really require all of these dependencies? I would recommend starting from a bare-bones Python installation and building an environment with only the packages necessary to run your application. Then, use that environment to create your environment.yml.

I put together an environment definition file that seems to work for working with GNNIS (at least the example in the README works) and that should be easier to install than the one that's currently in the repo (that one is really only useful if you are aiming tor reproduce exactly what Paul did).

We will eventually get something like this in the repo (after @katzberger is back from vacation), but for now you should hopefully be able to get a working environment if you save this locally and create an environment from it, and then install GNNIS either from a locally checked out clone of this repo or by doing python -m pip install git+https://github.com/rinikerlab/GNNImplicitSolvent

name: GNNImplicitSolvent_run
channels:
  - conda-forge
dependencies:
  - ipython
  - python=3.10
  - rdkit
  - pytorch
  - ca-certificates
  - openssl
  - scikit-learn
  - openff-toolkit
  - openmmforcefields
  - certifi
  - openmm-torch=1.4
  - openmoltools
  - h5py
  - numba
  - pytorch_geometric
  - pytorch_cluster
  - pytorch_scatter
  - pytorch_sparse
  - seaborn
  - nglview
  - pyemma

[PatWalters]

Thanks, @greglandrum that worked. I had to make one tweak to environment.yml. I couldn't install pyemma with conda, but pip worked.

channels:
  - conda-forge
dependencies:
  - ipython
  - python=3.10
  - rdkit
  - pytorch
  - ca-certificates
  - openssl
  - scikit-learn
  - openff-toolkit
  - openmmforcefields
  - certifi
  - openmm-torch=1.4
  - openmoltools
  - h5py
  - numba
  - pytorch_geometric
  - pytorch_cluster
  - pytorch_scatter
  - pytorch_sparse
  - seaborn
  - nglview
  - pip:
    - pyemma