Marker Intensity Predictor

This repo contains different part of Marker Intensity Research, based on HTAN data files.

It includes:

1. Jupyter notebooks
2. A linear regression model to calculate LR baseline data
3. Different auto encoders to train models
4. A plotting package to combine the results of linear regression and different auto encoders

Jupyter notebooks

Run the following to start the jupyter notebook:

$ virtualenv .venv && source .venv/bin/activate && python -m pip install -r requirements.txt && python -m ipykernel install --user --name=.venv
$ jupyter labextension install jupyterlab-plotly@4.14.2
$ jupyter-lab

Deep learning model training

Setup & Configuration

1. Create a virtual environment
2. pip install -r requirements.txt will install all required packages

ML Flow Integration

MLFlow is being used for experiment tracking.
Although a results folder is being used locally, all data is being stored using the backend of MLFlow.
Do NOT rely on the local results folder, it is only temporary.

Usage

python VAE/main.py model --file data/HTA9-3_Bx2_HMS_Tumor_quant.csv

Available Arguments

Long	Short	Description	Required
--file	x	The file to use for training the models	✓
--run	-r	The name of the run	✓
--experiment	-e	The name of the associated experiment. If no experiment with this name exists and new one will be created. Default: Default experiment
--morph	x	Should morphological features be included for training and evaluation? Default: true
--mode	x	Possible values: none, ae, vae. If ae or vae is selected only the respective model is being trained. No comparison will be done. Default: none.
--tracking_url	-t	The tracking url for the mlflow tracking server. Points to localhost as default. Please specify a valid url. Example: http:127.0.0.1:5000

Latent space exploration

A tool to explore the latent space of a VAE is included.
The tool is using Streamlit and MLFlow to provide interactive as well as tracking functionalities.