franek18/liquidbiopsycancerclassification

Liquid Biopsy Cancer Classification

This is the official code repository to the paper Platelet RNA sequencing data through the lens of machine learning.

Used libraries

Python==3.8.10
torch==1.7.1
torchvision==0.8.2
scikit-learn==0.24.1

How to start

First create Python3 virtual environment:

python3 -m venv venv

Then install all required Python packages:

pip install -r requirements.txt

The main file is cnn.py. In this file we must set such variables like:

• lr_list - list with at least one value of learning rate for our model
• Dropout_list - list with at least one value of Dropout of final layer for our model
• wd_list - list with at least one value of weight decay for optimizer of our model
• train_filename - name of numpy file with indices to the train samples from whole dataset. If doesn't exist it will be created a new file, with new train/test split.
• test_filename - name of numpy file with indices to the test samples from whole dataset. If doesn't exist it will be created a new file, with new train/test split.
• result_file - name of csv file for saving results like val and test auc for each fold.
• log_file - name of txt file for logs from training of a model.

In the same file, in function train we must set variable data_dir and annotation_file to the locations of dataset and file with annotations. It should be equal:

'annotations/Cancer_annotations_mts.csv'

which is set by function annotateCancer.