Using a Genetic Algorithm to Identify Co-morbidities

This github project is for the NICHD’s Decoding Maternal Morbidity Data Challenge.

It uses a novel genetic algorithm to identify co-morbidities associated with a particular outcome (here maternal morbidities), by optimizing the risk ratio given different categorical inputs.

Use notes:

• To be replicated, all scripts need to be run from the root folder
• The csv data for the nuMoM2b study (not uploaded to github), needs to be located at the data/nuMoM2b.csv to replicate the findings.
• This uses python 3.7, and the libraries specified in requirements.txt. See that file for notes on creating your own conda environment to replicate.
• The genetic algorithm does have stochastic elements, so your results may differ slightly from my results

The folder /tech_docs has more technical documentation on the genetic algorithm, but also note the source code for the functions is 100% provided in /src/genrules.py.

The jupyter notebook Example1_genrules.ipynb provides examples of the base algorithm to identify particular comorbidities. To run this notebook locally, you can use the command:

jupyter nbconvert --to notebook --execute Example1_genrules.ipynb --output Example1_genrules.ipynb

Or if you prefer to browse html output, you could use

jupyter nbconvert --execute Example1_genrules.ipynb --to html

The submission forms are located in the submission_forms folder (the registration and overall submission form).

If you have any questions, please feel free to contact me,

• Andrew Wheeler, PhD
• awheeler29@gsu.edu
• https://andrewpwheeler.com/