The Cancer Genome Atlas Program Project (692R) ⚛️

This is a repository dedicated to the course project of building an genome autoencoder for cancers. The autoencoder implementation was inspired by this example construction.

Contents 📦

scripts/: A collection of Bash scripts for various purposes. Tested on Bash v3 (macOS) and v4 (Linux).

• dataInfo.sh: Runs basic data analysis and generates log files containing the results.
• init.sh: Initialization script. MUST BE RUN FIRST (see the "Instructions" section below).
• masterFrame.sh: Requires completion of trimFile.sh in order to finish data cleaning and generate a single TSV dataframe.
• nullInfo.sh: Requires the log files from dataInfo.sh to perform basic missing value analysis.
• nullToZero.sh: Requires completion of masterFrame.sh so to replace all missing entries with zero.
• trimFile.sh: Basic cleaning of the data.

See the individual files for more detailed documentation.

autoencoder/: The autoencoder source files and configuration file.

• classes.py: Model architecture classes and the dataframe class.
• cnfg.json: Configuration file (see the "Instructions" section below).
• main.py: Autoencoder implementation. Loads the requisite model architecture from classes.py specified in cnfg.json.

Instructions 📝

To begin clone (or fork) this repository and place the TCGAdata.zip file in the project root. Then run ./scripts/init.sh to recursively extract the relevant contents to the newly instantiated data/ directory.

Now, in order to prep data/ for the autoencoder you will need to run the following.

  ./scripts/trimFile.sh && ./scripts/masterFrame.sh && ./scripts/nullToZero.sh

Before running the autoencoder, you will have to specify the desired configuration in cnfg.json. Here is the schema definition.

{
  "batch": 0, // (Integer) Batch size.
  "epochs": 0, // (Integer) Number of epochs.
  "load": null, // (null | String) A file path to a training checkpoint, which will be loaded as the model starting point.
  "model": "", // (String) The model to use; must be (a class name) from `classes.py` corresponding to a supported model.
  "save": null, // (null | String) A file path to where a checkpoint will be saved upon completion.
  "tsv": "" // (String) The file path to the TSV dataframe generated earlier.
}

Finally, to run the autoencoder simply invoke the following. Keep in mind you may need to install the requisite Python packages. Also, be conscience of the machine's capabilities; this is a resource heavy program!

  python3 autoencoder/main.py

Problems 🔴

If you encounter a problem see Issues. If someone has not already filed an issue for your problem, then please create one. Be descriptive.

Contributions 👥

Most welcome! Take a peak at Issues to see what is needed, or fork & PR with your desired improvement.