Skip to content

Latest commit

 

History

History
86 lines (76 loc) · 3.41 KB

README.md

File metadata and controls

86 lines (76 loc) · 3.41 KB
Raw

FlowAI

AI project for Flow Cytometry Analysis

This project is aimed at classifying flow cytometry data from 116 patients as either CLL (Chronic Lymphocytic Leukemia) or non-CLL using various Machine Learning methods. Data is provided by the James A. Haley VA in collaboration with Drs. Andrew Borkowski, Phil Foulis, and Loveleen Kang.

Prerequisites and Setup

These steps have been tested on macOS 10.15 (Catalina) Make sure you have python and pip installed: For Mac, first install homebrew and then

brew install python3

Also make sure you have virtualenv installed: 

python3 -m pip install --user virtualenv

Next clone the repo:

git clone https://github.com/kangakum36/FlowAI.git

Make a new virtual environment and activate it

cd FlowAI
python3 -m venv FlowAI
source FlowAI/bin/activate

Next install requirements

pip3 install -r requirements.txt

Finally, you have to install libomp for xgboost to work, on Mac this is:

brew install libomp

Data Download

Currently, the only way to get access to the data is by emailing me (kangakum [at] gmail [dot] com) for access to a gcp storage bucket. Although we have deidentified the data so it contains no information about the patients, we are trying to be extremely careful in regulating access.

Once you have access, install the google cloud sdk by following these instructions.
Make sure to say yes when prompted to add gcloud to your PATH, as the data download script won't work otherwise.
Then run 

gcloud init

Select the flowai project when prompted, you should already have access if you are at this step. Next run

gcloud components update

Finally, run the following commands (you must have gcloud added to your PATH) 

chmod +x ./scripts/download_data.sh
mkdir data
cd scripts
./download_data.sh

Making a new model

  1. In the Models folder, create a new file (e.g. <your_model_name>.py). Create a class for your model that inherits BaseModel.
  2. Override the train and test functions.
  3. In main.py, add an abbreviation for your model to the choices for valid abbreviations in the parse_args method.
  4. In the main method, there is a series of cascading if statements. Add your model to this series and do any necessary data processing and training.

Testing a model

To test a model on the data, 

python3 -W ignore main.py -m [model tag]

You should see the an output similar to the following

Selected model accuracy is: 0.875

Current model tags are 'xgb': Gradient Boosted RF Classifier, 'dtc': Decision Tree Classifier, 'nn': Vanilla Neural Net, 'cnn': Convolutional NN

Once you train/test a model once, the processed dataset will be placed in a pickle (.p) file in the data folder.
Using the -p flag on subsequent model tests will load the dataset from the saved file like so:

python3 -W ignore main.py -m [model tag] -p load

This way you don't have to wait for the data to be processed every time

Results and Remarks

Currently, we have implemented cross validation for the XGB and 1D-Convolutional Neural Network Models.
Additionally, we have grid-searched hyperparameters for the XGB model, as it was the most promising in initial tests.
Best average accuracy: Gradient Boosted Classifier (XGB) - 80% (F1 Score also 0.8)