CATH Hierrarchy Data

Data is sourced from: https://github.com/wouterboomsma/cath_datasets?tab=readme-ov-file

cath_3class_ca data:

• CATH database context and additional information on training data

Running Experiments

1. Activate the venv (w/ uv)

• Run uv venv to initialize the venv
• Activate the venv by with the following CLI commands:
  • MacOS: Run source .venv/bin/activate
  • Windows Run .venv\Scripts\activate.ps1
• Run uv sync to download the dependencies

2. Run the mlflow server

• Run mlflow ui
• Open your localhost at the specified port to let mlflow track the runs

3. Run the Python module

• Run python -m src.main
• Four models will be trained, evaluated, and logged (training time is based on an M1 Macbook Pro)
  1. NN - 1 Hidden Layer, 4 Neurons (~6s)
  2. NN - 2 Hidden Layers, 64 Neurons Each (~10s)
  3. CNN (<21min on my M1 Macbook Pro)
  4. Simplifed CNN w/ Early Stopping (<10min)

Modules

• src.models
  • Different experimental deep learning models using the keras/mlflow API
• src.utils
  • Interface for downloading/interacting with CATH protein dataset
  • Utilities for data pre- and post-processing
• src.main
  • Main entry point for experiment tracking