COVID Antibody Engineering

Engineering to improve the affinity of a COVID antibody using the dataset from Engelhart et al. 2021.

• See Data Exploration.ipynb for EDA.
• See Fitness Modeling.ipynb for antibody fitness modeling w.r.t binding affinity.
• See Candidate Generation.ipynb for generating candidate sequences for another round of wet-lab testing.
  • Proposed sequences are in data/ as explained in the notebook.

Install

• Install docker and start the daemon.

• docker-compose up will start the container and print URLs to connect to the Jupyter web UI.

  • If you're using VSCode, install the Remote Development Extensions Pack to easily run code and otherwise work within the docker container.
  • Attach to the {repo_name}-notebook-1 container.
  • Be sure to install the VSCode Python extension in the remote container as well.

• The Prefect UI is available at http://localhost:4200/ to view jobs.

• Jupyter notebooks are available at http://localhost:8888/.

  • Ensure that notebooks are "trusted" in the top right corner to display interactive plots.

• Download data: docker-compose exec -i notebook python /code/src/download_data.py

Usage

You can use poetry to run up an environment without Docker to save on overhead.

• Install poetry.
• Configure to create a virtual environment in the project: poetry config virtualenvs.in-project true`
• poetry install and point your notebooks to the venv.
• To add/remove packages:
  • poetry {add|remove} {package}
  • poetry export -f requirements.txt --output requirements.txt --without-hashes.
  • Run an environment shell: poetry shell
  • Run mflow locally: mlflow ui
  • Run docker-compose up --build to rebuild the image with new packages, or for faster turnaround, docker-compose exec notebook pip install -r /usr/src/app/requirements.txt
• To preprocess data for use in notebooks, run the Prefect flows from the ETL.py module:
  • extract_seeds_of_interest will preprocess some of the seed chains into dataframes of CDR mutations.
  • export_ESM_embeddings will preprocess the ESM embeddings for the sequences in the dataset library.