Predict Physical and Chemical Conditions from a Genome

[tylerbarnum]

This PR enables the use of the models to predict growth conditions for bacteria and archaea. It does not yet allow the user to reproduce the development or evaluation of the models.

To quickly test: you should be able to run this and have a result printed and a file test.predictions.tsv produced:

git clone -b tyler-genomic-features https://github.com/cultivarium/predict-media-physicochemistry.git
cd predict-media-physicochemistry
python -m venv venv
source venv/bin/activate
pip install -r requirements
python src/predict_physicochemistry.py -c tests/test_data/GCA_000172155.1_ASM17215v1_genomic.fna.gz -p tests/test_data/GCA_000172155.1_ASM17215v1_protein.faa.gz -o test.predictions.tsv

As noted in the README, you can also check out the notebook notebooks/tutorial.ipynb to see how individual functions work. This could be a helpful way you and users to understand the code before diving into the codebase. In particular the code to compute traits from BacDive download will make a lot more sense when you see that the data is a wacky nested dictionary.

Yet to come:

• Code for model training and selection
• Notebooks where that code and other analyses currently live
• Documentation for the above
• Confidence intervals for predictions
• Unit tests
• Rename

Other considerations:

• The setup.py is incomplete. I had a error relating to a pandas package availability and am curious what you report.
• FYI models are stored aka "pickled" using the package joblib.
• I tried to use classes to provide organization but I welcome suggestions for better organization.
• This PR merges in dev, and we can merge into main once unit tests, etc. are done.

[knightjdr]

For @alexcritschristoph, change pip install -r requirements to pip install -r requirements.txt in the instructions above. Technically also for @tylerbarnum, but want to make sure Alex sees.

[knightjdr]

Needs a .gitignore. A suitable one is here: https://gist.github.com/knightjdr/dc89bd33a4b21546c15822034871580b.

[knightjdr]

Issues:

• The output file is test.predictions.tsv.predictions.tsv.
• After fixing the above, add test.predictions.tsv to .gitignore

[tylerbarnum]

@knightjdr Any strong feelings on bumping the python version closer to current? I'm on 3.8 because a package that I no longer use required it

[knightjdr]

@knightjdr Any strong feelings on bumping the python version closer to current? I'm on 3.8 because a package that I no longer use required it

I would specify it at >= 3.8. Not everyone will be using a newer version, so it's good to provide some flexibility here. Compatibility with newer versions shouldn't be an issue.

[tylerbarnum]

Adding these from Alex to "yet to come" (next PR):

• Remove the NaN's in the current tabular output of the program.
• Add the units to the output table, and specify the units clearly in the documentation.
• Rename to genome-opts.py and make it a pip installable package.