leojklarner / gauche Goto Github PK

Parent:

• splitting and scaling

Molecular property prediction

• loading benchamarks
• validating SMILES
• featurisation to fingerprints, fragments and fragprints

Protein ligand binding affinity

• loading benchmarks
• validating pdb/sdf files
• featurisation to graph-based features, interaction-based features
• check whether ligand extraction is correct by comparing extracted ligand IDs to those in PDBbind

Currently the python versioning of this project is unclear: the README has no info, the supplied conda env says python 3.7, while the internal setup.py says python>=3.8. Maybe this should be standardized? It looks like any python version should do as long as the dependencies are supported, no?

The buckminsterfullerene gif has stopped working in the README :)

I'd recommend splitting dependencies into groups like this

This makes it easier to keep track of what is need for testing, dev, docs, prod etc.

Relevant lines for setup.py

The Google colab links for the notebook tutorials have no instructions for installing dependencies.

• include additional variable for internal representation which is not overwritten upon featurisation

• load benchmark sets (such as pdbbind)
• add download and splitting of arbitrary pdb complexes
• cluster by ligand and protein similarity
• add PLEC fingerprints, BINANA+Vina features, RFScore features
• add explicit hydrogens during bond order augmentation

Apologies for the ad-like post. Figured this was the easiest way to bring it on your radar though. It's a 2-day event happening next week, and details are at https://ac-bo-hackathon.github.io/. It would be great to have a Gauche contribution! Feel free to close the issue at any point.

It may be worth investigating whether convolutional kernel networks [1] can be integrated as a GP graph kernel.

[1] Chen, D., Jacob, L. and Mairal, J., Convolutional kernel networks for graph-structured data. ICML, 2020.

First off, great work, this is a really cool package!

I've been playing with the graph representation inputs using graphein to a model building off of SIGP (some examples in your codebase call it GraphGP) and have been getting some really great performance out of it. However, I'm struggling to understand how to correctly save and then load the model back into memory for inference after training. If I save the state dict then re-init using that state dict, the model performs as if it had been randomly initialized. I also tried pickling the model (not the ideal solution) I get the following exception:

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[88], [line 4](vscode-notebook-cell:?execution_count=88&line=4)
      [1](vscode-notebook-cell:?execution_count=88&line=1) import pickle
      [3](vscode-notebook-cell:?execution_count=88&line=3) with open('model.pkl', 'wb') as file:
----> [4](vscode-notebook-cell:?execution_count=88&line=4)     pickle.dump(model, file)

RuntimeError: Pickling of "rdkit.Chem.rdchem.Atom" instances is not enabled (http://www.boost.org/libs/python/doc/v2/pickle.html)

I tried setting train_inputs to None before saving. This took care of the exception, however I'm back to the original issue where the model seems to be randomly initialized.

I was wondering if you had any guidance here, or if there was something in the docs that I missed. Thanks!

Hi, thanks for your work! I'm wondering how exactly do I install gauche as a library. It seems that the instructions in the README are only for installing dependencies. Meanwhile, when I do pip install git+https://github.com/leojklarner/gauche.git, I get an error

error: subprocess-exited-with-error

  × Getting requirements to build wheel did not run successfully.
  │ exit code: 1
  ╰─> [1 lines of output]
      error in gauche setup command: 'extras_require' must be a dictionary whose values are strings or lists of strings containing valid project/version requirement specifiers.
      [end of output]

Thanks!

@a-r-j will address this in the next release of graphein I believe. Just leaving a TODO note here.

All dependencies mentioned in the readme are available as conda packages.
Always nicer to avoid pip when using anaconda imho. Also, much easier to prepare a yaml file?
you already have a requirements file. then you wouldn't need a setup.py file?
you know, a simple
Just a thought, not an issue per se.

I think moving the contents of the benchmarks directory into the codebase (.e.g gprotorch.benchmarks) will make organisation and docs clearer. It also helps (modestly) to enforce a consistent API across the library.

Also, we should rename from gprotorch to gauche.